enne2/qwentts.cpp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

tests logs

Browse Source
This commit is contained in:
Pascal
2026-05-11 16:27:12 +02:00
parent 32560ea597
commit eefb13043e
48 changed files with 3185 additions and 1119 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/base-CPU-BF16.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-base-BF16.gguf: 480 tensors, data at offset 5954496
[WeightCtx] Loaded 316 tensors, 3322.0 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 334.1 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[WeightCtx] Loaded 76 tensors, 34.9 MB into backend
[SpeakerEncoder] Loaded: enc_dim=2048 sr=24000 mel_dim=128 hidden=512 mfa=1536 asp_attn=128 se=128 scale=8
[GGUF] ../models/qwen-tokenizer-12hz-BF16.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=base tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=loaded speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=40, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [40, 2048] first4: 0.022157 -0.009049 0.008161 -0.020214
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.461393 -0.069925 0.016141 -0.039265
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.523327 -0.014488 0.346884 1.775712
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.294868 0.182233 0.448534 1.742519
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.770371 0.163638 0.800043 1.111360
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.047174 -34.401340 -7.069329 28.628973
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.239853 -1.222670 -0.223573 0.888168
[Debug] talker-logits-prefill: [3072] first4: -4.270525 -6.396639 -1.012693 -3.237629
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.042911 -0.033715 -0.000281 0.188977
[Debug] talker-hidden-step1: [2048] first4: 1.715209 -2.522261 6.795998 0.740671
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=310 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=679 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=896 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000016 0.000020 0.000017 0.000019
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-base-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/base/base-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 1.000000 max: 3.3516e-02 mean: 1.1754e-04
[Cossim] L7 cos: 0.999999 max: 1.2854e+01 mean: 1.5072e-03
[Cossim] L14 cos: 0.999999 max: 1.2859e+01 mean: 2.5488e-03
[Cossim] L21 cos: 0.999999 max: 1.2857e+01 mean: 6.7488e-03
[Cossim] L27 cos: 0.999999 max: 1.0791e+01 mean: 3.5090e-02
[Cossim] Final cos: 0.999997 max: 1.6618e-01 mean: 2.1030e-03
[Cossim] Logits cos: 0.999997 max: 4.6051e-02 mean: 7.1986e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.0431e-06 mean: 9.3024e-09
[Cossim] TalkerHiddenStep1 cos: 0.999979 max: 6.7536e-02 mean: 1.2194e-02
[Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 0.999625
[Cossim] WAV stft_cos: 0.999808 samples: 120960
tests/base-CPU-F32.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-base-F32.gguf: 480 tensors, data at offset 5954432
[WeightCtx] Loaded 316 tensors, 6643.5 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 668.1 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[WeightCtx] Loaded 76 tensors, 45.8 MB into backend
[SpeakerEncoder] Loaded: enc_dim=2048 sr=24000 mel_dim=128 hidden=512 mfa=1536 asp_attn=128 se=128 scale=8
[GGUF] ../models/qwen-tokenizer-12hz-F32.gguf: 398 tensors, data at offset 29664
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 116.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=base tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=loaded speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=40, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [40, 2048] first4: 0.022157 -0.009049 0.008161 -0.020214
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.460968 -0.069929 0.016046 -0.039092
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.525317 -0.007489 0.340217 1.777318
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.296772 0.189305 0.441262 1.743797
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.771753 0.171250 0.791419 1.111589
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.133091 -34.474182 -7.108906 28.674915
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.242892 -1.225839 -0.224931 0.890014
[Debug] talker-logits-prefill: [3072] first4: -4.261982 -6.399908 -0.996097 -3.238420
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.042911 -0.033715 -0.000281 0.188977
[Debug] talker-hidden-step1: [2048] first4: 1.683682 -2.524467 6.789284 0.745040
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=310 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=679 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=896 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000014 0.000016
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav
[Quant] F32 -> ../models/qwen-talker-1.7b-base-F32.gguf + ../models/qwen-tokenizer-12hz-F32.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-F32.gguf --codec ../models/qwen-tokenizer-12hz-F32.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/base/base-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 1.000000 max: 7.6294e-06 mean: 7.1945e-08
[Cossim] L7 cos: 1.000000 max: 2.1973e-03 mean: 4.1054e-07
[Cossim] L14 cos: 1.000000 max: 2.1973e-03 mean: 7.1001e-07
[Cossim] L21 cos: 1.000000 max: 2.1973e-03 mean: 1.9730e-06
[Cossim] L27 cos: 1.000000 max: 3.9062e-03 mean: 9.9707e-06
[Cossim] Final cos: 1.000000 max: 1.1396e-04 mean: 6.6050e-07
[Cossim] Logits cos: 1.000000 max: 2.1219e-05 mean: 3.0684e-06
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.0431e-06 mean: 9.3024e-09
[Cossim] TalkerHiddenStep1 cos: 1.000000 max: 1.5497e-05 mean: 2.0270e-06
[Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 0.999640
[Cossim] WAV stft_cos: 0.999815 samples: 120960
tests/base-CPU-Q4_K_M.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-base-Q4_K_M.gguf: 480 tensors, data at offset 5954496
[WeightCtx] Loaded 316 tensors, 1012.7 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 109.5 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[WeightCtx] Loaded 76 tensors, 34.9 MB into backend
[SpeakerEncoder] Loaded: enc_dim=2048 sr=24000 mel_dim=128 hidden=512 mfa=1536 asp_attn=128 se=128 scale=8
[GGUF] ../models/qwen-tokenizer-12hz-Q4_K_M.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 16.4 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 25.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 13.6 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=base tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=loaded speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=40, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [40, 2048] first4: 0.022105 -0.009079 0.008285 -0.019901
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005388 0.008927 -0.004415 0.000679
[Debug] tts-pad-embed: [2048] first4: -0.005388 0.008927 -0.004415 0.000679
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.441821 -0.069394 0.003658 0.020266
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -8.877851 0.790778 0.485222 0.194165
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -8.792478 1.086515 0.586075 0.265361
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -8.716181 1.630248 1.123896 -0.509474
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 4.479937 -33.404484 -6.656628 23.418547
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.151891 -1.182686 -0.209713 0.723736
[Debug] talker-logits-prefill: [3072] first4: -6.714985 -5.188290 0.217856 -2.100574
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=412 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=350 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1748 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1636 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1202 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=903 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=706 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1529 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 412.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.000769 -0.046681 -0.010753 0.124454
[Debug] talker-hidden-step1: [2048] first4: 2.581219 -2.674675 6.272708 -0.891046
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1134 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1176 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=947 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=819 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1202 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=993 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1870 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=743 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=903 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1759 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1845 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 412.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000015 0.000012 0.000014
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav
[Quant] Q4_K_M -> ../models/qwen-talker-1.7b-base-Q4_K_M.gguf + ../models/qwen-tokenizer-12hz-Q4_K_M.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-Q4_K_M.gguf --codec ../models/qwen-tokenizer-12hz-Q4_K_M.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/base/base-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 0.999695 max: 1.0566e-02 mean: 9.6975e-04
[Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] L0 cos: 0.999229 max: 9.8534e-01 mean: 6.6525e-03
[Cossim] L7 cos: 0.997455 max: 3.4714e+01 mean: 5.8715e-01
[Cossim] L14 cos: 0.997448 max: 3.4809e+01 mean: 6.7599e-01
[Cossim] L21 cos: 0.997233 max: 3.7314e+01 mean: 1.1012e+00
[Cossim] L27 cos: 0.976128 max: 2.2716e+03 mean: 5.1773e+00
[Cossim] Final cos: 0.945574 max: 2.2531e+01 mean: 2.9617e-01
[Cossim] Logits cos: 0.976703 max: 3.6042e+00 mean: 7.1212e-01
[Cossim] NextEmbStep0 cos: 0.804895 max: 2.9351e-01 mean: 3.5512e-02
[Cossim] TalkerHiddenStep1 cos: 0.737229 max: 7.2906e+00 mean: 1.3695e+00
[Cossim] CodesFull exact: 43.25% (1008 values)
[Cossim] Audio cos: 0.824962
[Cossim] WAV stft_cos: 0.884114 samples: 120960
tests/base-CPU-Q8_0.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-base-Q8_0.gguf: 480 tensors, data at offset 5954496
[WeightCtx] Loaded 316 tensors, 1765.0 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 177.5 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[WeightCtx] Loaded 76 tensors, 34.9 MB into backend
[SpeakerEncoder] Loaded: enc_dim=2048 sr=24000 mel_dim=128 hidden=512 mfa=1536 asp_attn=128 se=128 scale=8
[GGUF] ../models/qwen-tokenizer-12hz-Q8_0.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 30.9 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 33.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 25.6 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=base tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=loaded speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=40, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [40, 2048] first4: 0.021979 -0.009018 0.008292 -0.020170
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005270 0.008716 -0.004355 0.001036
[Debug] tts-pad-embed: [2048] first4: -0.005270 0.008716 -0.004355 0.001036
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.468042 -0.069505 0.014260 -0.037197
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -2.188824 0.154519 0.209425 1.591020
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.960747 0.344378 0.321142 1.566153
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -1.448369 0.302417 0.673400 0.974501
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 6.696815 -34.333122 -7.302668 28.583275
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.227877 -1.219968 -0.230900 0.886548
[Debug] talker-logits-prefill: [3072] first4: -4.138302 -6.346890 -1.057953 -3.235421
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.042839 -0.033702 -0.000574 0.188203
[Debug] talker-hidden-step1: [2048] first4: 1.741904 -2.502983 6.887070 0.710767
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=310 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=679 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=896 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000015 0.000017
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav
[Quant] Q8_0 -> ../models/qwen-talker-1.7b-base-Q8_0.gguf + ../models/qwen-tokenizer-12hz-Q8_0.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-Q8_0.gguf --codec ../models/qwen-tokenizer-12hz-Q8_0.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/base/base-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 0.999987 max: 2.2577e-03 mean: 1.8282e-04
[Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] L0 cos: 0.999987 max: 9.4385e-02 mean: 8.7919e-04
[Cossim] L7 cos: 0.999988 max: 2.5666e+00 mean: 4.3442e-02
[Cossim] L14 cos: 0.999987 max: 2.5645e+00 mean: 5.1566e-02
[Cossim] L21 cos: 0.999986 max: 2.5640e+00 mean: 8.9808e-02
[Cossim] L27 cos: 0.999896 max: 1.0790e+02 mean: 4.3627e-01
[Cossim] Final cos: 0.998976 max: 4.7232e+00 mean: 2.9227e-02
[Cossim] Logits cos: 0.999922 max: 2.6035e-01 mean: 3.8184e-02
[Cossim] NextEmbStep0 cos: 0.999992 max: 1.1117e-03 mean: 2.2935e-04
[Cossim] TalkerHiddenStep1 cos: 0.999585 max: 2.9974e-01 mean: 5.4856e-02
[Cossim] CodesFull exact: 99.21% (1008 values)
[Cossim] Audio cos: 0.999640
[Cossim] WAV stft_cos: 0.999815 samples: 120960
tests/base-CUDA0-BF16.log
+96 -1
View File
@@ -18,7 +18,85 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512 [Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend [WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[CausalTransConv] FATAL: 'tok_dec.upsample.0.conv.weight' expected F32 or F16, got type 30 [WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=base tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=loaded speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=40, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [40, 2048] first4: 0.022157 -0.009049 0.008161 -0.020214
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.461242 -0.069840 0.015607 -0.039501
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.533321 -0.007802 0.339130 1.795481
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.305173 0.189369 0.439706 1.762431
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.781766 0.171444 0.791870 1.132004
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.054309 -34.616005 -7.153164 28.758896
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.239884 -1.229216 -0.226025 0.891412
[Debug] talker-logits-prefill: [3072] first4: -4.187500 -6.406250 -1.031250 -3.234375
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.042911 -0.033715 -0.000281 0.188977
[Debug] talker-hidden-step1: [2048] first4: 1.700760 -2.524712 6.793815 0.719804
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=310 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=679 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=896 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000016 0.000020 0.000016 0.000017
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Lang: english Seed: 42 MaxNewTokens: 64 [Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy [Input] Mode: greedy
@@ -28,3 +106,20 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav [Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-base-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf [Quant] BF16 -> ../models/qwen-talker-1.7b-base-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy [GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/base/base-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.999998 max: 3.7647e-02 mean: 2.6221e-04
[Cossim] L7 cos: 0.999995 max: 3.0323e+01 mean: 3.1956e-03
[Cossim] L14 cos: 0.999995 max: 3.0341e+01 mean: 5.1099e-03
[Cossim] L21 cos: 0.999995 max: 3.0325e+01 mean: 1.3364e-02
[Cossim] L27 cos: 0.999988 max: 3.3269e+01 mean: 7.9881e-02
[Cossim] Final cos: 0.999982 max: 5.2641e-01 mean: 4.8446e-03
[Cossim] Logits cos: 0.999984 max: 1.1177e-01 mean: 1.7884e-02
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.0431e-06 mean: 9.3024e-09
[Cossim] TalkerHiddenStep1 cos: 0.999990 max: 5.7575e-02 mean: 8.6064e-03
[Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 0.981072
[Cossim] WAV stft_cos: 0.988772 samples: 120960
tests/base-CUDA0-F32.log
+26 -26
View File
@@ -20,15 +20,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend [WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7 [Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 199.8 MB into backend [WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 199.8 MB [DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 6.0 MB into backend [WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 48.2 MB into backend [WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 48.2 MB [SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend [WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000 [EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 4.0 MB into backend [WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 4.0 MB [EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend [WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB [EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz [Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
@@ -42,13 +42,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [40, 2048] first4: 0.022157 -0.009049 0.008161 -0.020214 [Debug] talker-input-embed: [40, 2048] first4: 0.022157 -0.009049 0.008161 -0.020214
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141 [Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141 [Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.460968 -0.069929 0.016047 -0.039092 [Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.460921 -0.069914 0.016044 -0.039100
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.525317 -0.007488 0.340215 1.777318 [Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.525369 -0.007542 0.340209 1.777390
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.296772 0.189306 0.441261 1.743797 [Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.296884 0.189210 0.441208 1.743850
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.771753 0.171251 0.791418 1.111589 [Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.771875 0.171125 0.791471 1.111795
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.133080 -34.474182 -7.108909 28.674915 [Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.131204 -34.471611 -7.107906 28.673965
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.242892 -1.225839 -0.224931 0.890014 [Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.242832 -1.225767 -0.224903 0.889998
[Debug] talker-logits-prefill: [3072] first4: -4.261972 -6.399906 -0.996095 -3.238420 [Debug] talker-logits-prefill: [3072] first4: -4.249419 -6.400744 -0.999631 -3.239202
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
@@ -67,7 +67,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.042911 -0.033715 -0.000281 0.188977 [Debug] next-emb-step0: [2048] first4: -0.042911 -0.033715 -0.000281 0.188977
[Debug] talker-hidden-step1: [2048] first4: 1.683677 -2.524468 6.789280 0.745043 [Debug] talker-hidden-step1: [2048] first4: 1.686512 -2.523739 6.789270 0.739157
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
@@ -94,7 +94,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000015 0.000017 [Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000014 0.000016
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08 [Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 1.000000 max: 7.6294e-06 mean: 7.2511e-08 [Cossim] L0 cos: 1.000000 max: 1.5526e-03 mean: 2.7874e-05
[Cossim] L7 cos: 1.000000 max: 4.8828e-04 mean: 2.9065e-07 [Cossim] L7 cos: 1.000000 max: 6.9824e-02 mean: 1.0704e-04
[Cossim] L14 cos: 1.000000 max: 4.8828e-04 mean: 6.4957e-07 [Cossim] L14 cos: 1.000000 max: 7.0312e-02 mean: 2.7487e-04
[Cossim] L21 cos: 1.000000 max: 4.8828e-04 mean: 1.9102e-06 [Cossim] L21 cos: 1.000000 max: 7.1777e-02 mean: 9.2039e-04
[Cossim] L27 cos: 1.000000 max: 2.6855e-03 mean: 7.3149e-06 [Cossim] L27 cos: 1.000000 max: 6.8652e-01 mean: 3.4051e-03
[Cossim] Final cos: 1.000000 max: 5.0545e-05 mean: 4.7584e-07 [Cossim] Final cos: 1.000000 max: 2.3113e-02 mean: 2.3500e-04
[Cossim] Logits cos: 1.000000 max: 1.5259e-05 mean: 2.1238e-06 [Cossim] Logits cos: 1.000000 max: 1.7665e-02 mean: 2.5834e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.0431e-06 mean: 9.3024e-09 [Cossim] NextEmbStep0 cos: 1.000000 max: 1.0431e-06 mean: 9.3024e-09
[Cossim] TalkerHiddenStep1 cos: 1.000000 max: 1.4544e-05 mean: 2.0600e-06 [Cossim] TalkerHiddenStep1 cos: 1.000000 max: 1.0001e-02 mean: 1.8443e-03
[Cossim] CodesFull exact: 100.00% (1008 values) [Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 0.981927 [Cossim] Audio cos: 0.981624
[Cossim] WAV stft_cos: 0.989278 samples: 120960 [Cossim] WAV stft_cos: 0.989090 samples: 120960
tests/base-CUDA0-Q4_K_M.log
+28 -28
View File
@@ -42,13 +42,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [40, 2048] first4: 0.022105 -0.009079 0.008285 -0.019901 [Debug] talker-input-embed: [40, 2048] first4: 0.022105 -0.009079 0.008285 -0.019901
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005388 0.008927 -0.004415 0.000679 [Debug] trailing-text-hidden: [1, 2048] first4: -0.005388 0.008927 -0.004415 0.000679
[Debug] tts-pad-embed: [2048] first4: -0.005388 0.008927 -0.004415 0.000679 [Debug] tts-pad-embed: [2048] first4: -0.005388 0.008927 -0.004415 0.000679
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.447358 -0.062733 0.006189 0.016085 [Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.446561 -0.060163 0.007940 0.018917
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -8.846175 0.797751 0.613486 0.111307 [Debug] talker-hidden-prefill-l7: [40, 2048] first4: -9.003338 0.695780 0.437512 -0.092255
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -8.782341 1.143654 0.709960 0.178828 [Debug] talker-hidden-prefill-l14: [40, 2048] first4: -8.947271 1.029292 0.524048 -0.026420
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -8.581591 1.646079 1.301754 -0.414389 [Debug] talker-hidden-prefill-l21: [40, 2048] first4: -8.766617 1.521604 1.084662 -0.644349
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 4.360051 -31.890827 -6.769986 22.778488 [Debug] talker-hidden-prefill-l27: [40, 2048] first4: 3.957858 -31.895205 -7.048655 22.182625
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.147967 -1.130170 -0.213488 0.704625 [Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.134484 -1.131720 -0.222550 0.687040
[Debug] talker-logits-prefill: [3072] first4: -6.952660 -4.840123 0.336555 -1.941904 [Debug] talker-logits-prefill: [3072] first4: -6.833356 -4.878899 0.348303 -1.763907
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=412 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=412 u=-1.0000000000 subseq=2
@@ -67,23 +67,23 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 412.000000 22.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 412.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.000769 -0.046681 -0.010753 0.124454 [Debug] next-emb-step0: [2048] first4: -0.000769 -0.046681 -0.010753 0.124454
[Debug] talker-hidden-step1: [2048] first4: 2.771324 -2.807498 6.536952 -0.717196 [Debug] talker-hidden-step1: [2048] first4: 2.841227 -2.793463 6.694829 -0.852604
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1134 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1134 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1176 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1176 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=947 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=947 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=819 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=819 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1085 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1202 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=108 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=993 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1870 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=455 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=743 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=743 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=903 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1110 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1759 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=781 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1845 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=82 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999695 max: 1.0566e-02 mean: 9.6975e-04 [Cossim] Embed cos: 0.999695 max: 1.0566e-02 mean: 9.6975e-04
[Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] L0 cos: 0.999266 max: 6.6433e-01 mean: 6.6418e-03 [Cossim] L0 cos: 0.999257 max: 6.4607e-01 mean: 6.6580e-03
[Cossim] L7 cos: 0.997458 max: 3.4222e+01 mean: 5.8657e-01 [Cossim] L7 cos: 0.997458 max: 3.4807e+01 mean: 5.8686e-01
[Cossim] L14 cos: 0.997451 max: 3.4328e+01 mean: 6.7567e-01 [Cossim] L14 cos: 0.997451 max: 3.4921e+01 mean: 6.7611e-01
[Cossim] L21 cos: 0.997227 max: 4.1874e+01 mean: 1.1036e+00 [Cossim] L21 cos: 0.997236 max: 4.1071e+01 mean: 1.0997e+00
[Cossim] L27 cos: 0.982909 max: 1.6701e+03 mean: 5.0607e+00 [Cossim] L27 cos: 0.983920 max: 1.5898e+03 mean: 5.0195e+00
[Cossim] Final cos: 0.944197 max: 2.4863e+01 mean: 2.9907e-01 [Cossim] Final cos: 0.944976 max: 2.4265e+01 mean: 2.9897e-01
[Cossim] Logits cos: 0.975393 max: 3.6027e+00 mean: 7.2304e-01 [Cossim] Logits cos: 0.974293 max: 3.5458e+00 mean: 7.4524e-01
[Cossim] NextEmbStep0 cos: 0.804895 max: 2.9351e-01 mean: 3.5512e-02 [Cossim] NextEmbStep0 cos: 0.804895 max: 2.9351e-01 mean: 3.5512e-02
[Cossim] TalkerHiddenStep1 cos: 0.745795 max: 7.8548e+00 mean: 1.3562e+00 [Cossim] TalkerHiddenStep1 cos: 0.744627 max: 8.0302e+00 mean: 1.3702e+00
[Cossim] CodesFull exact: 28.47% (1008 values) [Cossim] CodesFull exact: 28.08% (1008 values)
[Cossim] Audio cos: 0.790801 [Cossim] Audio cos: 0.773401
[Cossim] WAV stft_cos: 0.859112 samples: 120960 [Cossim] WAV stft_cos: 0.845611 samples: 120960
tests/base-CUDA0-Q8_0.log
+16 -16
View File
@@ -42,13 +42,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [40, 2048] first4: 0.021979 -0.009018 0.008292 -0.020170 [Debug] talker-input-embed: [40, 2048] first4: 0.021979 -0.009018 0.008292 -0.020170
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005270 0.008716 -0.004355 0.001036 [Debug] trailing-text-hidden: [1, 2048] first4: -0.005270 0.008716 -0.004355 0.001036
[Debug] tts-pad-embed: [2048] first4: -0.005270 0.008716 -0.004355 0.001036 [Debug] tts-pad-embed: [2048] first4: -0.005270 0.008716 -0.004355 0.001036
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.467419 -0.069494 0.014718 -0.037177 [Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.467628 -0.069474 0.014061 -0.036574
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -2.191055 0.153858 0.210238 1.590774 [Debug] talker-hidden-prefill-l7: [40, 2048] first4: -2.188927 0.157751 0.206575 1.589240
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.963035 0.342891 0.321307 1.565399 [Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.961075 0.345972 0.318182 1.565162
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -1.450296 0.306020 0.671899 0.975008 [Debug] talker-hidden-prefill-l21: [40, 2048] first4: -1.449745 0.302517 0.671301 0.977163
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 6.666507 -34.310875 -7.308170 28.604483 [Debug] talker-hidden-prefill-l27: [40, 2048] first4: 6.664756 -34.346832 -7.308811 28.629057
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.226686 -1.218321 -0.230912 0.886583 [Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.226849 -1.220797 -0.231159 0.888217
[Debug] talker-logits-prefill: [3072] first4: -4.066623 -6.434669 -1.084206 -3.275083 [Debug] talker-logits-prefill: [3072] first4: -4.188689 -6.376436 -1.055355 -3.192302
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
@@ -67,7 +67,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.042839 -0.033702 -0.000574 0.188203 [Debug] next-emb-step0: [2048] first4: -0.042839 -0.033702 -0.000574 0.188203
[Debug] talker-hidden-step1: [2048] first4: 1.614406 -2.553735 6.809586 0.754741 [Debug] talker-hidden-step1: [2048] first4: 1.656439 -2.512899 6.874210 0.605789
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999987 max: 2.2577e-03 mean: 1.8282e-04 [Cossim] Embed cos: 0.999987 max: 2.2577e-03 mean: 1.8282e-04
[Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] L0 cos: 0.999987 max: 9.8606e-02 mean: 8.8043e-04 [Cossim] L0 cos: 0.999987 max: 1.0142e-01 mean: 8.7943e-04
[Cossim] L7 cos: 0.999987 max: 5.0764e+00 mean: 4.3612e-02 [Cossim] L7 cos: 0.999987 max: 4.2681e+00 mean: 4.3422e-02
[Cossim] L14 cos: 0.999987 max: 5.0305e+00 mean: 5.1840e-02 [Cossim] L14 cos: 0.999987 max: 4.2227e+00 mean: 5.1565e-02
[Cossim] L21 cos: 0.999986 max: 4.8687e+00 mean: 8.9429e-02 [Cossim] L21 cos: 0.999986 max: 4.0930e+00 mean: 9.0034e-02
[Cossim] L27 cos: 0.999898 max: 1.1023e+02 mean: 4.2984e-01 [Cossim] L27 cos: 0.999891 max: 1.1262e+02 mean: 4.3764e-01
[Cossim] Final cos: 0.999054 max: 4.1020e+00 mean: 2.8603e-02 [Cossim] Final cos: 0.999029 max: 4.9770e+00 mean: 2.8865e-02
[Cossim] Logits cos: 0.999911 max: 2.4493e-01 mean: 3.8623e-02 [Cossim] Logits cos: 0.999943 max: 2.2792e-01 mean: 3.1810e-02
[Cossim] NextEmbStep0 cos: 0.999992 max: 1.1117e-03 mean: 2.2935e-04 [Cossim] NextEmbStep0 cos: 0.999992 max: 1.1117e-03 mean: 2.2935e-04
[Cossim] TalkerHiddenStep1 cos: 0.999728 max: 2.9192e-01 mean: 4.3826e-02 [Cossim] TalkerHiddenStep1 cos: 0.999703 max: 2.5879e-01 mean: 4.6164e-02
[Cossim] CodesFull exact: 99.90% (1008 values) [Cossim] CodesFull exact: 99.90% (1008 values)
[Cossim] Audio cos: 0.981566 [Cossim] Audio cos: 0.981566
[Cossim] WAV stft_cos: 0.989070 samples: 120960 [Cossim] WAV stft_cos: 0.989070 samples: 120960
tests/base-Vulkan0-BF16.log
+96 -1
View File
@@ -18,7 +18,85 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512 [Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend [WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[CausalTransConv] FATAL: 'tok_dec.upsample.0.conv.weight' expected F32 or F16, got type 30 [WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=base tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=loaded speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=40, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [40, 2048] first4: 0.022157 -0.009049 0.008161 -0.020214
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.461532 -0.069882 0.015988 -0.039216
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.523219 -0.014190 0.347239 1.776223
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.294983 0.182637 0.448559 1.742983
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.770002 0.164026 0.799983 1.111261
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.051283 -34.398327 -7.069520 28.629948
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.239994 -1.222570 -0.223580 0.888203
[Debug] talker-logits-prefill: [3072] first4: -4.302365 -6.420417 -0.993868 -3.248467
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.042911 -0.033715 -0.000281 0.188977
[Debug] talker-hidden-step1: [2048] first4: 1.675931 -2.521131 6.784448 0.760312
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=310 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=679 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=896 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000016 0.000020 0.000015 0.000017
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Lang: english Seed: 42 MaxNewTokens: 64 [Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy [Input] Mode: greedy
@@ -28,3 +106,20 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav [Python] Audio: 120960 samples 24000 Hz 5.04s -> python/base/base-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-base-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf [Quant] BF16 -> ../models/qwen-talker-1.7b-base-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy [GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-base-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --lang english --max-new 64 --dump cpp/base -o cpp/base/base-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/base/base-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.999999 max: 7.3315e-02 mean: 1.2200e-04
[Cossim] L7 cos: 0.999999 max: 1.2859e+01 mean: 1.5268e-03
[Cossim] L14 cos: 0.999999 max: 1.2861e+01 mean: 2.5770e-03
[Cossim] L21 cos: 0.999999 max: 1.2868e+01 mean: 6.8197e-03
[Cossim] L27 cos: 0.999999 max: 1.0757e+01 mean: 3.8101e-02
[Cossim] Final cos: 0.999996 max: 1.6886e-01 mean: 2.3693e-03
[Cossim] Logits cos: 0.999993 max: 7.7975e-02 mean: 9.9753e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.0431e-06 mean: 9.3024e-09
[Cossim] TalkerHiddenStep1 cos: 0.999994 max: 4.1780e-02 mean: 6.7959e-03
[Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 0.981091
[Cossim] WAV stft_cos: 0.988782 samples: 120960
tests/base-Vulkan0-F32.log
+60 -60
View File
@@ -20,15 +20,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend [WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7 [Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 199.8 MB into backend [WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 199.8 MB [DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 6.0 MB into backend [WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 48.2 MB into backend [WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 48.2 MB [SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend [WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000 [EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 4.0 MB into backend [WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 4.0 MB [EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend [WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB [EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz [Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
@@ -43,47 +43,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141 [Debug] trailing-text-hidden: [1, 2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141 [Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.457824 -0.069581 0.015775 -0.038616 [Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.457824 -0.069581 0.015775 -0.038616
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.527061 -0.012803 0.350664 1.772303 [Debug] talker-hidden-prefill-l7: [40, 2048] first4: -1.524411 -0.013424 0.350418 1.770865
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.298302 0.183709 0.452202 1.738756 [Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.295802 0.183276 0.451901 1.737421
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.772566 0.165500 0.805204 1.105845 [Debug] talker-hidden-prefill-l21: [40, 2048] first4: -0.769769 0.164919 0.805717 1.104015
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.084506 -34.605457 -7.071080 28.531052 [Debug] talker-hidden-prefill-l27: [40, 2048] first4: 7.088632 -34.574314 -7.067035 28.496386
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.241432 -1.231496 -0.223914 0.886260 [Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.241578 -1.230414 -0.223791 0.885203
[Debug] talker-logits-prefill: [3072] first4: -4.187500 -6.421875 -1.005859 -3.253906 [Debug] talker-logits-prefill: [3072] first4: -4.246094 -6.417969 -0.998535 -3.251953
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1853 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=369 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=889 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1274 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=338 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=373 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=359 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=651 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1572 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1160 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1079 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=780 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 887.000000 1853.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.019854 -0.049802 -0.034545 0.027441 [Debug] next-emb-step0: [2048] first4: -0.042911 -0.033715 -0.000281 0.188977
[Debug] talker-hidden-step1: [2048] first4: -1.350096 -4.437189 1.978907 -1.633949 [Debug] talker-hidden-step1: [2048] first4: 1.656970 -2.523548 6.764471 0.742943
[Sample] step=1 c0=1028 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=113 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1705 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1357 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=310 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=496 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1461 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=926 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=318 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1275 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=525 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=679 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1842 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1748 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=896 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1918 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=889 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1670 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=207 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -93,8 +93,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 887.000000 1853.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: -0.000264 -0.000364 -0.000392 -0.000405 [Debug] output-audio: [122880] first4: 0.000015 0.000018 0.000016 0.000016
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08 [Cossim] Embed cos: 1.000000 max: 2.2650e-06 mean: 1.8841e-08
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.999998 max: 7.0589e-02 mean: 2.3487e-04 [Cossim] L0 cos: 0.999998 max: 7.0589e-02 mean: 2.3855e-04
[Cossim] L7 cos: 0.999997 max: 2.0187e+01 mean: 2.8197e-03 [Cossim] L7 cos: 0.999998 max: 1.6196e+01 mean: 2.8051e-03
[Cossim] L14 cos: 0.999997 max: 2.0179e+01 mean: 4.5399e-03 [Cossim] L14 cos: 0.999998 max: 1.6191e+01 mean: 4.6960e-03
[Cossim] L21 cos: 0.999997 max: 2.0261e+01 mean: 1.2459e-02 [Cossim] L21 cos: 0.999997 max: 1.6287e+01 mean: 1.2335e-02
[Cossim] L27 cos: 0.999991 max: 3.5026e+01 mean: 8.5269e-02 [Cossim] L27 cos: 0.999994 max: 2.8662e+01 mean: 7.8427e-02
[Cossim] Final cos: 0.999982 max: 4.8414e-01 mean: 4.7583e-03 [Cossim] Final cos: 0.999984 max: 4.3231e-01 mean: 4.5308e-03
[Cossim] Logits cos: 0.999993 max: 7.4465e-02 mean: 1.1664e-02 [Cossim] Logits cos: 0.999996 max: 5.2906e-02 mean: 8.9718e-03
[Cossim] NextEmbStep0 cos: 0.431734 max: 5.4164e-01 mean: 5.7355e-02 [Cossim] NextEmbStep0 cos: 1.000000 max: 1.0431e-06 mean: 9.3024e-09
[Cossim] TalkerHiddenStep1 cos: 0.282685 max: 2.3768e+01 mean: 2.1941e+00 [Cossim] TalkerHiddenStep1 cos: 0.999988 max: 7.7208e-02 mean: 9.3441e-03
[Cossim] CodesFull exact: 0.30% (1008 values) [Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 0.039381 [Cossim] Audio cos: 0.981673
[Cossim] WAV stft_cos: 0.057063 samples: 120960 [Cossim] WAV stft_cos: 0.989114 samples: 120960
tests/base-Vulkan0-Q4_K_M.log
+52 -52
View File
@@ -44,46 +44,46 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] tts-pad-embed: [2048] first4: -0.005388 0.008927 -0.004415 0.000679 [Debug] tts-pad-embed: [2048] first4: -0.005388 0.008927 -0.004415 0.000679
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.440114 -0.067513 0.007682 0.014812 [Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.440114 -0.067513 0.007682 0.014812
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -8.839585 0.783318 0.508945 0.097937 [Debug] talker-hidden-prefill-l7: [40, 2048] first4: -8.839585 0.783318 0.508945 0.097937
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -8.774442 1.117288 0.612085 0.163537 [Debug] talker-hidden-prefill-l14: [40, 2048] first4: -8.774428 1.117173 0.612158 0.163505
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -8.658731 1.690545 1.112828 -0.787406 [Debug] talker-hidden-prefill-l21: [40, 2048] first4: -8.657841 1.690515 1.112808 -0.787705
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 4.962493 -31.760586 -6.788569 21.825624 [Debug] talker-hidden-prefill-l27: [40, 2048] first4: 4.949123 -31.748377 -6.797402 21.811562
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.167658 -1.120514 -0.213115 0.672126 [Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.167212 -1.120122 -0.213400 0.671716
[Debug] talker-logits-prefill: [3072] first4: -6.832031 -4.953125 0.424805 -1.972656 [Debug] talker-logits-prefill: [3072] first4: -6.800781 -4.984375 0.422607 -1.982422
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=412 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1853 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1294 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=350 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=778 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1748 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1206 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=1636 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=308 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1202 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1117 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1439 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=359 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=55 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=903 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=453 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=706 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=989 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=1529 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=403 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=363 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 887.000000 1853.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 412.000000 22.000000
[Debug] next-emb-step0: [2048] first4: 0.016646 -0.005029 -0.033803 -0.002747 [Debug] next-emb-step0: [2048] first4: -0.000769 -0.046681 -0.010753 0.124454
[Debug] talker-hidden-step1: [2048] first4: -0.929883 -3.736696 0.872852 -1.414843 [Debug] talker-hidden-step1: [2048] first4: 2.759117 -2.726088 6.474637 -0.801468
[Sample] step=1 c0=1014 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=780 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1134 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1203 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1315 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1176 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=77 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=947 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=111 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=819 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=592 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1202 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1932 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=993 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=813 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1870 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=1506 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=743 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=547 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=903 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=2001 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=111 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1759 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1942 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=1845 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=1285 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -93,8 +93,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 887.000000 1853.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 412.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000014 0.000013 0.000006 0.000011 [Debug] output-audio: [122880] first4: 0.000014 0.000014 0.000012 0.000014
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999695 max: 1.0566e-02 mean: 9.6975e-04 [Cossim] Embed cos: 0.999695 max: 1.0566e-02 mean: 9.6975e-04
[Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] L0 cos: 0.999279 max: 7.9302e-01 mean: 6.4647e-03 [Cossim] L0 cos: 0.999278 max: 8.0864e-01 mean: 6.4649e-03
[Cossim] L7 cos: 0.997448 max: 4.1743e+01 mean: 5.8622e-01 [Cossim] L7 cos: 0.997447 max: 4.5748e+01 mean: 5.8610e-01
[Cossim] L14 cos: 0.997441 max: 4.3301e+01 mean: 6.7443e-01 [Cossim] L14 cos: 0.997441 max: 4.7305e+01 mean: 6.7424e-01
[Cossim] L21 cos: 0.997231 max: 4.2807e+01 mean: 1.0945e+00 [Cossim] L21 cos: 0.997231 max: 4.6822e+01 mean: 1.0947e+00
[Cossim] L27 cos: 0.982530 max: 1.7326e+03 mean: 5.0100e+00 [Cossim] L27 cos: 0.982515 max: 1.7328e+03 mean: 5.0134e+00
[Cossim] Final cos: 0.946231 max: 2.4114e+01 mean: 2.9488e-01 [Cossim] Final cos: 0.946228 max: 2.4165e+01 mean: 2.9492e-01
[Cossim] Logits cos: 0.976299 max: 3.4270e+00 mean: 7.1229e-01 [Cossim] Logits cos: 0.976501 max: 3.4192e+00 mean: 7.1027e-01
[Cossim] NextEmbStep0 cos: 0.439674 max: 4.9057e-01 mean: 5.7374e-02 [Cossim] NextEmbStep0 cos: 0.804895 max: 2.9351e-01 mean: 3.5512e-02
[Cossim] TalkerHiddenStep1 cos: 0.265089 max: 2.4202e+01 mean: 2.2974e+00 [Cossim] TalkerHiddenStep1 cos: 0.745895 max: 7.7025e+00 mean: 1.3559e+00
[Cossim] CodesFull exact: 0.20% (1008 values) [Cossim] CodesFull exact: 25.69% (1008 values)
[Cossim] Audio cos: 0.274084 [Cossim] Audio cos: 0.785361
[Cossim] WAV stft_cos: 0.412402 samples: 120960 [Cossim] WAV stft_cos: 0.854673 samples: 120960
tests/base-Vulkan0-Q8_0.log
+53 -53
View File
@@ -43,47 +43,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] trailing-text-hidden: [1, 2048] first4: -0.005270 0.008716 -0.004355 0.001036 [Debug] trailing-text-hidden: [1, 2048] first4: -0.005270 0.008716 -0.004355 0.001036
[Debug] tts-pad-embed: [2048] first4: -0.005270 0.008716 -0.004355 0.001036 [Debug] tts-pad-embed: [2048] first4: -0.005270 0.008716 -0.004355 0.001036
[Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.461420 -0.072044 0.014792 -0.037443 [Debug] talker-hidden-prefill-l0: [40, 2048] first4: -0.461420 -0.072044 0.014792 -0.037443
[Debug] talker-hidden-prefill-l7: [40, 2048] first4: -2.199468 0.166968 0.200645 1.604671 [Debug] talker-hidden-prefill-l7: [40, 2048] first4: -2.199481 0.165310 0.198761 1.609799
[Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.975133 0.353360 0.306994 1.579719 [Debug] talker-hidden-prefill-l14: [40, 2048] first4: -1.975275 0.352142 0.305376 1.584287
[Debug] talker-hidden-prefill-l21: [40, 2048] first4: -1.461743 0.289562 0.665961 0.999581 [Debug] talker-hidden-prefill-l21: [40, 2048] first4: -1.463363 0.287608 0.663962 1.005549
[Debug] talker-hidden-prefill-l27: [40, 2048] first4: 6.705354 -34.567703 -7.414961 28.435642 [Debug] talker-hidden-prefill-l27: [40, 2048] first4: 6.687571 -34.492264 -7.387871 28.361349
[Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.228340 -1.229235 -0.234628 0.882638 [Debug] talker-hidden-prefill-final: [40, 2048] first4: 0.227624 -1.225957 -0.233658 0.879905
[Debug] talker-logits-prefill: [3072] first4: -4.078125 -6.410156 -1.056641 -3.250000 [Debug] talker-logits-prefill: [3072] first4: -4.097656 -6.390625 -1.061523 -3.240234
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1853 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=369 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=889 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1274 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=338 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=373 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1795 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=651 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1572 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1160 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1079 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=780 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 887.000000 1853.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.011231 -0.048261 -0.014040 0.004942 [Debug] next-emb-step0: [2048] first4: -0.042839 -0.033702 -0.000574 0.188203
[Debug] talker-hidden-step1: [2048] first4: -1.172883 -4.615955 1.940985 -1.469731 [Debug] talker-hidden-step1: [2048] first4: 1.651582 -2.566358 6.810423 0.678816
[Sample] step=1 c0=1028 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1763 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1722 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1203 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1114 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=310 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1041 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1461 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1595 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1445 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=930 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=87 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1144 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=1146 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=679 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=478 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=474 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=896 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1654 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=889 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=14 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=777 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -93,8 +93,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 887.000000 1853.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: -0.000002 -0.000010 -0.000003 -0.000009 [Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000015 0.000017
[WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/base/base-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/base/base-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999987 max: 2.2577e-03 mean: 1.8282e-04 [Cossim] Embed cos: 0.999987 max: 2.2577e-03 mean: 1.8282e-04
[Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] L0 cos: 0.999991 max: 9.1096e-02 mean: 6.4066e-04 [Cossim] L0 cos: 0.999992 max: 9.1096e-02 mean: 6.4077e-04
[Cossim] L7 cos: 0.999986 max: 2.1562e+01 mean: 4.2083e-02 [Cossim] L7 cos: 0.999986 max: 2.1572e+01 mean: 4.2006e-02
[Cossim] L14 cos: 0.999986 max: 2.1610e+01 mean: 4.8769e-02 [Cossim] L14 cos: 0.999986 max: 2.1626e+01 mean: 4.8583e-02
[Cossim] L21 cos: 0.999985 max: 2.1766e+01 mean: 7.8347e-02 [Cossim] L21 cos: 0.999985 max: 2.1762e+01 mean: 7.8263e-02
[Cossim] L27 cos: 0.999957 max: 5.4678e+01 mean: 3.4737e-01 [Cossim] L27 cos: 0.999955 max: 5.8713e+01 mean: 3.4944e-01
[Cossim] Final cos: 0.999642 max: 4.1274e+00 mean: 2.2081e-02 [Cossim] Final cos: 0.999624 max: 4.1470e+00 mean: 2.2311e-02
[Cossim] Logits cos: 0.999946 max: 2.0047e-01 mean: 3.0143e-02 [Cossim] Logits cos: 0.999949 max: 1.8228e-01 mean: 2.9513e-02
[Cossim] NextEmbStep0 cos: 0.426517 max: 5.3731e-01 mean: 5.7581e-02 [Cossim] NextEmbStep0 cos: 0.999992 max: 1.1117e-03 mean: 2.2935e-04
[Cossim] TalkerHiddenStep1 cos: 0.292946 max: 2.1366e+01 mean: 2.2204e+00 [Cossim] TalkerHiddenStep1 cos: 0.999821 max: 1.7147e-01 mean: 3.5799e-02
[Cossim] CodesFull exact: 0.20% (1008 values) [Cossim] CodesFull exact: 99.90% (1008 values)
[Cossim] Audio cos: 0.077502 [Cossim] Audio cos: 0.981612
[Cossim] WAV stft_cos: 0.115858 samples: 120960 [Cossim] WAV stft_cos: 0.989094 samples: 120960
tests/clone-CPU-BF16.log
+12 -12
View File
@@ -75,7 +75,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.770371 0.163638 0.800043 1.111360 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.770371 0.163638 0.800043 1.111360
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.047174 -34.401340 -7.069329 28.628973 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.047174 -34.401340 -7.069329 28.628973
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.239853 -1.222670 -0.223573 0.888168 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.239853 -1.222670 -0.223573 0.888168
[Debug] talker-logits-prefill: [3072] first4: 1.343600 -7.634940 -6.992353 -10.519046 [Debug] talker-logits-prefill: [3072] first4: 1.354070 -7.634824 -6.983518 -10.512266
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
@@ -94,7 +94,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=365 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=365 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.073561 -0.034233 -0.008814 0.042004 [Debug] next-emb-step0: [2048] first4: 0.073561 -0.034233 -0.008814 0.042004
[Debug] talker-hidden-step1: [2048] first4: 0.566472 2.581657 -0.032918 0.227096 [Debug] talker-hidden-step1: [2048] first4: 0.563787 2.589628 -0.021490 0.225289
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
@@ -146,14 +146,14 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.989590 max: 7.7399e-01 mean: 8.9716e-03 [Cossim] L0 cos: 0.989590 max: 7.7399e-01 mean: 8.9716e-03
[Cossim] L7 cos: 0.999993 max: 1.2854e+01 mean: 2.3280e-02 [Cossim] L7 cos: 0.999993 max: 1.2854e+01 mean: 2.3288e-02
[Cossim] L14 cos: 0.999976 max: 1.2858e+01 mean: 5.2397e-02 [Cossim] L14 cos: 0.999976 max: 1.2858e+01 mean: 5.2396e-02
[Cossim] L21 cos: 0.999696 max: 1.7395e+01 mean: 2.0170e-01 [Cossim] L21 cos: 0.999696 max: 1.7455e+01 mean: 2.0176e-01
[Cossim] L27 cos: 0.998838 max: 4.6131e+01 mean: 4.6407e-01 [Cossim] L27 cos: 0.998836 max: 4.6225e+01 mean: 4.6439e-01
[Cossim] Final cos: 0.993945 max: 1.9927e+01 mean: 1.5086e-01 [Cossim] Final cos: 0.993938 max: 1.9965e+01 mean: 1.5095e-01
[Cossim] Logits cos: 0.999804 max: 5.7617e-01 mean: 7.8600e-02 [Cossim] Logits cos: 0.999809 max: 5.5566e-01 mean: 7.7546e-02
[Cossim] NextEmbStep0 cos: 0.967766 max: 6.3431e-02 mean: 1.2632e-02 [Cossim] NextEmbStep0 cos: 0.967766 max: 6.3431e-02 mean: 1.2632e-02
[Cossim] TalkerHiddenStep1 cos: 0.998228 max: 7.7749e-01 mean: 1.3022e-01 [Cossim] TalkerHiddenStep1 cos: 0.998229 max: 7.7430e-01 mean: 1.3003e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 0.999979 max: 6.8943e-02 mean: 1.0444e-03 [Cossim] SpkMFA cos: 0.999979 max: 6.8943e-02 mean: 1.0444e-03
[Cossim] SpkASP cos: 0.999999 max: 2.2756e-03 mean: 1.8859e-04 [Cossim] SpkASP cos: 0.999999 max: 2.2756e-03 mean: 1.8859e-04
[Cossim] SpeakerEmb cos: 1.000000 max: 6.4747e-04 mean: 1.3471e-04 [Cossim] SpeakerEmb cos: 1.000000 max: 6.4747e-04 mean: 1.3471e-04
[Cossim] CodesFull exact: 3.77% (1008 values) [Cossim] CodesFull exact: 3.57% (1008 values)
[Cossim] Audio cos: -0.002504 [Cossim] Audio cos: 0.000390
[Cossim] WAV stft_cos: 0.222491 samples: 120960 [Cossim] WAV stft_cos: 0.161882 samples: 120960
tests/clone-CPU-F32.log
+4 -4
View File
@@ -75,7 +75,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.771753 0.171250 0.791419 1.111589 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.771753 0.171250 0.791419 1.111589
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.133091 -34.474182 -7.108906 28.674915 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.133091 -34.474182 -7.108906 28.674915
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.242892 -1.225839 -0.224931 0.890014 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.242892 -1.225839 -0.224931 0.890014
[Debug] talker-logits-prefill: [3072] first4: 1.330122 -7.768506 -6.935927 -10.338604 [Debug] talker-logits-prefill: [3072] first4: 1.330122 -7.768506 -6.935926 -10.338602
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
@@ -94,7 +94,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=365 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=365 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.073561 -0.034233 -0.008814 0.042004 [Debug] next-emb-step0: [2048] first4: 0.073561 -0.034233 -0.008814 0.042004
[Debug] talker-hidden-step1: [2048] first4: 0.584089 2.683546 -0.065922 0.135429 [Debug] talker-hidden-step1: [2048] first4: 0.584086 2.683545 -0.065921 0.135426
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
@@ -149,9 +149,9 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] L7 cos: 0.999998 max: 1.2205e+00 mean: 7.8975e-03 [Cossim] L7 cos: 0.999998 max: 1.2205e+00 mean: 7.8975e-03
[Cossim] L14 cos: 0.999994 max: 2.2472e+00 mean: 2.0513e-02 [Cossim] L14 cos: 0.999994 max: 2.2472e+00 mean: 2.0513e-02
[Cossim] L21 cos: 0.999920 max: 1.3672e+01 mean: 8.9238e-02 [Cossim] L21 cos: 0.999920 max: 1.3672e+01 mean: 8.9238e-02
[Cossim] L27 cos: 0.999724 max: 2.5152e+01 mean: 2.0175e-01 [Cossim] L27 cos: 0.999724 max: 2.5153e+01 mean: 2.0175e-01
[Cossim] Final cos: 0.998553 max: 1.1353e+01 mean: 6.5865e-02 [Cossim] Final cos: 0.998553 max: 1.1353e+01 mean: 6.5865e-02
[Cossim] Logits cos: 0.999966 max: 2.0577e-01 mean: 3.0349e-02 [Cossim] Logits cos: 0.999966 max: 2.0577e-01 mean: 3.0348e-02
[Cossim] NextEmbStep0 cos: 0.967766 max: 6.3431e-02 mean: 1.2632e-02 [Cossim] NextEmbStep0 cos: 0.967766 max: 6.3431e-02 mean: 1.2632e-02
[Cossim] TalkerHiddenStep1 cos: 0.999113 max: 4.3387e-01 mean: 9.2055e-02 [Cossim] TalkerHiddenStep1 cos: 0.999113 max: 4.3387e-01 mean: 9.2055e-02
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
tests/clone-CPU-Q4_K_M.log
+33 -33
View File
@@ -75,7 +75,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -8.716181 1.630248 1.123896 -0.509474 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -8.716181 1.630248 1.123896 -0.509474
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 4.479937 -33.404484 -6.656628 23.418547 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 4.479937 -33.404484 -6.656628 23.418547
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.151891 -1.182686 -0.209713 0.723736 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.151891 -1.182686 -0.209713 0.723736
[Debug] talker-logits-prefill: [3072] first4: -0.049402 -4.614957 -7.224392 -10.561956 [Debug] talker-logits-prefill: [3072] first4: 0.085887 -4.716971 -7.420881 -10.257107
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
@@ -84,33 +84,33 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=4 c=1412 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1412 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1155 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=1155 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=815 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=815 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=106 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1004 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=153 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=153 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1776 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=1376 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1435 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=790 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1728 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=208 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1922 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=1682 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1022 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1816 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=302 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=302 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 417.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 417.000000
[Debug] next-emb-step0: [2048] first4: 0.033382 0.011384 -0.022313 0.047956 [Debug] next-emb-step0: [2048] first4: 0.061441 0.001319 -0.046349 0.063087
[Debug] talker-hidden-step1: [2048] first4: 0.727982 2.735055 1.014960 0.314809 [Debug] talker-hidden-step1: [2048] first4: 0.986520 2.459286 1.038020 0.591026
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1210 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=405 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=985 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=390 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1636 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=447 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=329 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1059 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=564 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=2030 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1004 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=437 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=1887 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=706 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1713 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1529 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1484 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=85 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=82 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=1435 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -121,7 +121,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 417.000000 [Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 417.000000
[Debug] output-audio: [122880] first4: -0.001169 -0.002026 -0.001938 -0.000989 [Debug] output-audio: [122880] first4: -0.001189 -0.002031 -0.001900 -0.000914
[WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.922425 max: 2.2247e-01 mean: 2.1113e-02 [Cossim] Embed cos: 0.922425 max: 2.2247e-01 mean: 2.1113e-02
[Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] L0 cos: 0.963454 max: 9.8532e-01 mean: 2.7584e-02 [Cossim] L0 cos: 0.963454 max: 9.8533e-01 mean: 2.7584e-02
[Cossim] L7 cos: 0.997433 max: 3.4714e+01 mean: 1.6242e-01 [Cossim] L7 cos: 0.997433 max: 3.4717e+01 mean: 1.6236e-01
[Cossim] L14 cos: 0.997361 max: 3.4809e+01 mean: 2.4571e-01 [Cossim] L14 cos: 0.997361 max: 3.4811e+01 mean: 2.4537e-01
[Cossim] L21 cos: 0.995869 max: 3.6073e+01 mean: 7.7571e-01 [Cossim] L21 cos: 0.995884 max: 3.6484e+01 mean: 7.7258e-01
[Cossim] L27 cos: 0.945093 max: 2.1444e+03 mean: 2.0117e+00 [Cossim] L27 cos: 0.943101 max: 2.1955e+03 mean: 2.0073e+00
[Cossim] Final cos: 0.954199 max: 5.7803e+01 mean: 5.5407e-01 [Cossim] Final cos: 0.954425 max: 5.6869e+01 mean: 5.5228e-01
[Cossim] Logits cos: 0.991186 max: 3.6044e+00 mean: 5.1607e-01 [Cossim] Logits cos: 0.992545 max: 3.3069e+00 mean: 4.6488e-01
[Cossim] NextEmbStep0 cos: 0.828841 max: 1.4047e-01 mean: 2.9099e-02 [Cossim] NextEmbStep0 cos: 0.815309 max: 1.4610e-01 mean: 3.0946e-02
[Cossim] TalkerHiddenStep1 cos: 0.963338 max: 3.9269e+00 mean: 5.8955e-01 [Cossim] TalkerHiddenStep1 cos: 0.960525 max: 4.3099e+00 mean: 6.0835e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 9.8418e-03 mean: 1.3212e-04 [Cossim] SpkMFA cos: 1.000000 max: 9.8418e-03 mean: 1.3212e-04
[Cossim] SpkASP cos: 1.000000 max: 2.6596e-04 mean: 2.3921e-05 [Cossim] SpkASP cos: 1.000000 max: 2.6596e-04 mean: 2.3921e-05
[Cossim] SpeakerEmb cos: 1.000000 max: 2.3174e-04 mean: 1.6094e-05 [Cossim] SpeakerEmb cos: 1.000000 max: 2.3174e-04 mean: 1.6094e-05
[Cossim] CodesFull exact: 2.08% (1008 values) [Cossim] CodesFull exact: 1.59% (1008 values)
[Cossim] Audio cos: -0.004291 [Cossim] Audio cos: -0.016422
[Cossim] WAV stft_cos: 0.191699 samples: 120960 [Cossim] WAV stft_cos: 0.232122 samples: 120960
tests/clone-CPU-Q8_0.log
+27 -27
View File
@@ -75,26 +75,26 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -1.448369 0.302417 0.673400 0.974501 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -1.448369 0.302417 0.673400 0.974501
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 6.696815 -34.333122 -7.302668 28.583275 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 6.696815 -34.333122 -7.302668 28.583275
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.227877 -1.219968 -0.230900 0.886548 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.227877 -1.219968 -0.230900 0.886548
[Debug] talker-logits-prefill: [3072] first4: 1.200291 -7.925805 -6.852166 -10.537476 [Debug] talker-logits-prefill: [3072] first4: 1.240433 -7.908016 -6.891863 -10.518490
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=400 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=400 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=550 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=240 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1506 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=2010 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1192 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=196 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1162 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=580 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=853 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1713 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=692 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1221 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=18 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=1796 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1502 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1228 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1702 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=2015 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1860 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=143 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1008 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1062 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=265 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=1094 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.033802 -0.026094 0.058828 0.088771 [Debug] next-emb-step0: [2048] first4: 0.010604 -0.008920 0.074910 0.036289
[Debug] talker-hidden-step1: [2048] first4: 0.394369 2.910929 -0.614081 0.194712 [Debug] talker-hidden-step1: [2048] first4: 0.081502 2.761908 -0.857437 0.129533
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
@@ -121,7 +121,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000 [Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: -0.001312 -0.002237 -0.002049 -0.000977 [Debug] output-audio: [122880] first4: -0.001370 -0.002378 -0.002224 -0.001049
[WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -146,14 +146,14 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] L0 cos: 0.992177 max: 7.7281e-01 mean: 7.3119e-03 [Cossim] L0 cos: 0.992177 max: 7.7281e-01 mean: 7.3119e-03
[Cossim] L7 cos: 0.999983 max: 2.5666e+00 mean: 2.6623e-02 [Cossim] L7 cos: 0.999983 max: 2.5669e+00 mean: 2.6620e-02
[Cossim] L14 cos: 0.999969 max: 2.5645e+00 mean: 5.2457e-02 [Cossim] L14 cos: 0.999970 max: 2.5645e+00 mean: 5.2341e-02
[Cossim] L21 cos: 0.999711 max: 1.7818e+01 mean: 2.0121e-01 [Cossim] L21 cos: 0.999712 max: 1.7699e+01 mean: 2.0014e-01
[Cossim] L27 cos: 0.998660 max: 1.1899e+02 mean: 4.8494e-01 [Cossim] L27 cos: 0.998671 max: 9.6115e+01 mean: 4.8216e-01
[Cossim] Final cos: 0.994302 max: 1.6731e+01 mean: 1.4969e-01 [Cossim] Final cos: 0.994343 max: 1.6736e+01 mean: 1.4889e-01
[Cossim] Logits cos: 0.999849 max: 6.0111e-01 mean: 6.8404e-02 [Cossim] Logits cos: 0.999851 max: 6.0750e-01 mean: 6.3832e-02
[Cossim] NextEmbStep0 cos: 0.825702 max: 2.4852e-01 mean: 2.9797e-02 [Cossim] NextEmbStep0 cos: 0.829032 max: 1.5491e-01 mean: 2.9164e-02
[Cossim] TalkerHiddenStep1 cos: 0.988163 max: 2.2708e+00 mean: 3.3062e-01 [Cossim] TalkerHiddenStep1 cos: 0.993362 max: 2.1273e+00 mean: 2.4912e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 9.8418e-03 mean: 1.3212e-04 [Cossim] SpkMFA cos: 1.000000 max: 9.8418e-03 mean: 1.3212e-04
[Cossim] SpkASP cos: 1.000000 max: 2.6596e-04 mean: 2.3921e-05 [Cossim] SpkASP cos: 1.000000 max: 2.6596e-04 mean: 2.3921e-05
[Cossim] SpeakerEmb cos: 1.000000 max: 2.3174e-04 mean: 1.6094e-05 [Cossim] SpeakerEmb cos: 1.000000 max: 2.3174e-04 mean: 1.6094e-05
[Cossim] CodesFull exact: 2.68% (1008 values) [Cossim] CodesFull exact: 3.08% (1008 values)
[Cossim] Audio cos: -0.002931 [Cossim] Audio cos: -0.015470
[Cossim] WAV stft_cos: 0.145943 samples: 120960 [Cossim] WAV stft_cos: 0.196771 samples: 120960
tests/clone-CUDA0-BF16.log
+13 -13
View File
@@ -75,7 +75,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.770047 0.168826 0.790043 1.108689 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.770047 0.168826 0.790043 1.108689
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.103000 -34.495621 -7.125972 28.862335 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.103000 -34.495621 -7.125972 28.862335
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.241600 -1.225247 -0.225222 0.894842 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.241600 -1.225247 -0.225222 0.894842
[Debug] talker-logits-prefill: [3072] first4: 0.847656 -7.875000 -6.593750 -10.375000 [Debug] talker-logits-prefill: [3072] first4: 0.871094 -7.843750 -6.593750 -10.375000
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
@@ -94,7 +94,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=1094 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=1094 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.010340 -0.009316 0.074921 0.036954 [Debug] next-emb-step0: [2048] first4: 0.010340 -0.009316 0.074921 0.036954
[Debug] talker-hidden-step1: [2048] first4: 0.028097 2.930421 -0.771587 0.041852 [Debug] talker-hidden-step1: [2048] first4: 0.011413 2.945210 -0.770962 0.047850
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.965579 max: 1.9437e-01 mean: 9.8515e-03 [Cossim] Embed cos: 0.965579 max: 1.9437e-01 mean: 9.8515e-03
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.984331 max: 8.0261e-01 mean: 1.3008e-02 [Cossim] L0 cos: 0.984330 max: 8.0271e-01 mean: 1.3008e-02
[Cossim] L7 cos: 0.999981 max: 4.7529e+01 mean: 3.1958e-02 [Cossim] L7 cos: 0.999981 max: 4.7524e+01 mean: 3.1968e-02
[Cossim] L14 cos: 0.999956 max: 4.7530e+01 mean: 7.0140e-02 [Cossim] L14 cos: 0.999956 max: 4.7527e+01 mean: 7.0172e-02
[Cossim] L21 cos: 0.999568 max: 4.7582e+01 mean: 2.6741e-01 [Cossim] L21 cos: 0.999568 max: 4.7592e+01 mean: 2.6749e-01
[Cossim] L27 cos: 0.998450 max: 4.7913e+01 mean: 6.0463e-01 [Cossim] L27 cos: 0.998447 max: 4.7917e+01 mean: 6.0538e-01
[Cossim] Final cos: 0.992009 max: 1.5263e+01 mean: 1.9648e-01 [Cossim] Final cos: 0.992000 max: 1.5222e+01 mean: 1.9659e-01
[Cossim] Logits cos: 0.999456 max: 1.0356e+00 mean: 1.3013e-01 [Cossim] Logits cos: 0.999476 max: 1.0044e+00 mean: 1.2385e-01
[Cossim] NextEmbStep0 cos: 0.829096 max: 1.5468e-01 mean: 2.9154e-02 [Cossim] NextEmbStep0 cos: 0.829096 max: 1.5468e-01 mean: 2.9154e-02
[Cossim] TalkerHiddenStep1 cos: 0.992008 max: 2.2751e+00 mean: 2.7376e-01 [Cossim] TalkerHiddenStep1 cos: 0.991905 max: 2.2751e+00 mean: 2.7539e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 0.999973 max: 7.5789e-02 mean: 1.1867e-03 [Cossim] SpkMFA cos: 0.999973 max: 7.5789e-02 mean: 1.1867e-03
[Cossim] SpkASP cos: 0.999999 max: 3.9836e-03 mean: 2.4063e-04 [Cossim] SpkASP cos: 0.999999 max: 3.9836e-03 mean: 2.4063e-04
[Cossim] SpeakerEmb cos: 1.000000 max: 1.7091e-03 mean: 2.1396e-04 [Cossim] SpeakerEmb cos: 1.000000 max: 1.7091e-03 mean: 2.1396e-04
[Cossim] CodesFull exact: 4.76% (1008 values) [Cossim] CodesFull exact: 2.68% (1008 values)
[Cossim] Audio cos: 0.062271 [Cossim] Audio cos: 0.042042
[Cossim] WAV stft_cos: 0.382412 samples: 120960 [Cossim] WAV stft_cos: 0.253914 samples: 120960
tests/clone-CUDA0-F32.log
+19 -19
View File
@@ -69,13 +69,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141 [Debug] tts-pad-embed: [2048] first4: -0.005331 0.008751 -0.004225 0.001141
[Debug] speaker-emb: [2048] first4: 0.098317 -0.023761 -0.011860 0.006760 [Debug] speaker-emb: [2048] first4: 0.098317 -0.023761 -0.011860 0.006760
[Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000 [Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000
[Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.460968 -0.069929 0.016046 -0.039092 [Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.460921 -0.069914 0.016044 -0.039100
[Debug] talker-hidden-prefill-l7: [225, 2048] first4: -1.525316 -0.007488 0.340215 1.777316 [Debug] talker-hidden-prefill-l7: [225, 2048] first4: -1.525366 -0.007545 0.340203 1.777391
[Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.296772 0.189306 0.441261 1.743796 [Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.296881 0.189208 0.441204 1.743854
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.771752 0.171251 0.791418 1.111587 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.771864 0.171112 0.791441 1.111791
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.133099 -34.474205 -7.108918 28.674948 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.131247 -34.471676 -7.107930 28.674025
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.242893 -1.225840 -0.224932 0.890015 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.242833 -1.225769 -0.224904 0.890000
[Debug] talker-logits-prefill: [3072] first4: 1.172282 -7.792083 -6.821526 -10.286682 [Debug] talker-logits-prefill: [3072] first4: 1.173973 -7.794936 -6.823968 -10.287047
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
@@ -94,7 +94,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=541 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=541 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.003672 0.000908 0.039506 0.094594 [Debug] next-emb-step0: [2048] first4: 0.003672 0.000908 0.039506 0.094594
[Debug] talker-hidden-step1: [2048] first4: 0.753700 2.407002 -0.055373 0.158864 [Debug] talker-hidden-step1: [2048] first4: 0.755211 2.408826 -0.056213 0.159002
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.972420 max: 1.7531e-01 mean: 8.3100e-03 [Cossim] Embed cos: 0.972420 max: 1.7531e-01 mean: 8.3100e-03
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.986912 max: 6.6438e-01 mean: 1.1101e-02 [Cossim] L0 cos: 0.986912 max: 6.6446e-01 mean: 1.1103e-02
[Cossim] L7 cos: 0.999992 max: 1.2571e+00 mean: 2.7507e-02 [Cossim] L7 cos: 0.999992 max: 1.2571e+00 mean: 2.7510e-02
[Cossim] L14 cos: 0.999973 max: 2.1778e+00 mean: 5.9922e-02 [Cossim] L14 cos: 0.999973 max: 2.1783e+00 mean: 5.9936e-02
[Cossim] L21 cos: 0.999719 max: 8.4742e+00 mean: 2.2263e-01 [Cossim] L21 cos: 0.999719 max: 8.3895e+00 mean: 2.2266e-01
[Cossim] L27 cos: 0.998970 max: 2.4310e+01 mean: 5.1154e-01 [Cossim] L27 cos: 0.998971 max: 2.4050e+01 mean: 5.1167e-01
[Cossim] Final cos: 0.994696 max: 7.3989e+00 mean: 1.6647e-01 [Cossim] Final cos: 0.994705 max: 7.3234e+00 mean: 1.6647e-01
[Cossim] Logits cos: 0.999745 max: 5.7267e-01 mean: 8.8663e-02 [Cossim] Logits cos: 0.999743 max: 5.7313e-01 mean: 8.8918e-02
[Cossim] NextEmbStep0 cos: 0.829216 max: 1.2523e-01 mean: 2.9129e-02 [Cossim] NextEmbStep0 cos: 0.829216 max: 1.2523e-01 mean: 2.9129e-02
[Cossim] TalkerHiddenStep1 cos: 0.993850 max: 1.4298e+00 mean: 2.4206e-01 [Cossim] TalkerHiddenStep1 cos: 0.993853 max: 1.4289e+00 mean: 2.4205e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 3.8069e-04 mean: 1.3195e-06 [Cossim] SpkMFA cos: 1.000000 max: 3.8069e-04 mean: 1.3195e-06
[Cossim] SpkASP cos: 1.000000 max: 7.9870e-06 mean: 2.2269e-07 [Cossim] SpkASP cos: 1.000000 max: 7.9870e-06 mean: 2.2269e-07
[Cossim] SpeakerEmb cos: 1.000000 max: 1.4305e-05 mean: 2.5463e-07 [Cossim] SpeakerEmb cos: 1.000000 max: 1.4305e-05 mean: 2.5463e-07
[Cossim] CodesFull exact: 3.27% (1008 values) [Cossim] CodesFull exact: 3.17% (1008 values)
[Cossim] Audio cos: 0.053931 [Cossim] Audio cos: 0.042680
[Cossim] WAV stft_cos: 0.375521 samples: 120960 [Cossim] WAV stft_cos: 0.347584 samples: 120960
tests/clone-CUDA0-Q4_K_M.log
+38 -38
View File
@@ -69,13 +69,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] tts-pad-embed: [2048] first4: -0.005388 0.008927 -0.004415 0.000679 [Debug] tts-pad-embed: [2048] first4: -0.005388 0.008927 -0.004415 0.000679
[Debug] speaker-emb: [2048] first4: 0.098318 -0.023761 -0.011860 0.006760 [Debug] speaker-emb: [2048] first4: 0.098318 -0.023761 -0.011860 0.006760
[Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 696.000000 1522.000000 [Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 696.000000 1522.000000
[Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.447360 -0.062734 0.006191 0.016084 [Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.446565 -0.060165 0.007938 0.018915
[Debug] talker-hidden-prefill-l7: [225, 2048] first4: -8.839272 0.815075 0.612550 0.113270 [Debug] talker-hidden-prefill-l7: [225, 2048] first4: -8.827673 0.817571 0.615257 0.113759
[Debug] talker-hidden-prefill-l14: [225, 2048] first4: -8.772307 1.161852 0.711070 0.177234 [Debug] talker-hidden-prefill-l14: [225, 2048] first4: -8.766668 1.160747 0.712421 0.181608
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -8.581620 1.664241 1.294096 -0.437955 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -8.576015 1.672577 1.300637 -0.409118
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 4.385426 -32.019020 -6.841087 22.716158 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 4.231728 -31.911999 -6.799148 22.786421
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.148700 -1.133733 -0.215544 0.702090 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.143801 -1.132405 -0.214689 0.705796
[Debug] talker-logits-prefill: [3072] first4: -0.447498 -5.802912 -7.130526 -10.586733 [Debug] talker-logits-prefill: [3072] first4: -0.637080 -5.627232 -7.042767 -10.283505
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1534 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1534 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1233 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1233 u=-1.0000000000 subseq=2
@@ -88,29 +88,29 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=8 c=153 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=153 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=958 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=958 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1904 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1904 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=127 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1100 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1682 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=161 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=58 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1385 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=816 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=19 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1534.000000 1233.000000 362.000000 [Debug] codes-step0: [16] first4: 27.000000 1534.000000 1233.000000 362.000000
[Debug] next-emb-step0: [2048] first4: 0.045909 0.017556 -0.023983 0.005290 [Debug] next-emb-step0: [2048] first4: 0.027166 0.013352 -0.030278 0.042965
[Debug] talker-hidden-step1: [2048] first4: 0.116329 2.313672 0.636690 0.456164 [Debug] talker-hidden-step1: [2048] first4: 0.186110 2.303282 0.609568 0.448766
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=494 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=933 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=2 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=405 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=985 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=390 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=849 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=447 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1345 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1059 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1300 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=564 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=2030 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1004 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1162 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=903 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=729 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1713 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=344 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1484 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=85 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=85 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=496 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=1435 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -121,7 +121,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1534.000000 1233.000000 362.000000 [Debug] codes-full: [64, 16] first4: 27.000000 1534.000000 1233.000000 362.000000
[Debug] output-audio: [122880] first4: -0.001337 -0.002394 -0.002279 -0.001187 [Debug] output-audio: [122880] first4: -0.001325 -0.002367 -0.002474 -0.001618
[WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.918419 max: 1.9968e-01 mean: 2.1849e-02 [Cossim] Embed cos: 0.918419 max: 1.9968e-01 mean: 2.1849e-02
[Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] L0 cos: 0.961553 max: 8.6603e-01 mean: 2.8456e-02 [Cossim] L0 cos: 0.961565 max: 8.6245e-01 mean: 2.8455e-02
[Cossim] L7 cos: 0.997438 max: 3.4831e+01 mean: 1.6350e-01 [Cossim] L7 cos: 0.997436 max: 3.4796e+01 mean: 1.6355e-01
[Cossim] L14 cos: 0.997362 max: 3.4932e+01 mean: 2.4923e-01 [Cossim] L14 cos: 0.997360 max: 3.4896e+01 mean: 2.4907e-01
[Cossim] L21 cos: 0.995759 max: 4.1386e+01 mean: 7.9421e-01 [Cossim] L21 cos: 0.995756 max: 4.2230e+01 mean: 7.9445e-01
[Cossim] L27 cos: 0.958854 max: 1.6261e+03 mean: 2.0433e+00 [Cossim] L27 cos: 0.957830 max: 1.6528e+03 mean: 2.0496e+00
[Cossim] Final cos: 0.951848 max: 3.2548e+01 mean: 5.7157e-01 [Cossim] Final cos: 0.951977 max: 3.0445e+01 mean: 5.7143e-01
[Cossim] Logits cos: 0.990380 max: 3.5209e+00 mean: 5.1124e-01 [Cossim] Logits cos: 0.991294 max: 3.3204e+00 mean: 4.8721e-01
[Cossim] NextEmbStep0 cos: 0.796671 max: 1.4999e-01 mean: 3.2483e-02 [Cossim] NextEmbStep0 cos: 0.795254 max: 1.3611e-01 mean: 3.2456e-02
[Cossim] TalkerHiddenStep1 cos: 0.956649 max: 4.6231e+00 mean: 6.2956e-01 [Cossim] TalkerHiddenStep1 cos: 0.960882 max: 5.0472e+00 mean: 5.9496e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 3.8069e-04 mean: 1.3207e-06 [Cossim] SpkMFA cos: 1.000000 max: 3.8069e-04 mean: 1.3207e-06
[Cossim] SpkASP cos: 1.000000 max: 7.8678e-06 mean: 2.2438e-07 [Cossim] SpkASP cos: 1.000000 max: 7.8678e-06 mean: 2.2438e-07
[Cossim] SpeakerEmb cos: 1.000000 max: 1.4305e-05 mean: 2.5522e-07 [Cossim] SpeakerEmb cos: 1.000000 max: 1.4305e-05 mean: 2.5522e-07
[Cossim] CodesFull exact: 1.09% (1008 values) [Cossim] CodesFull exact: 1.69% (1008 values)
[Cossim] Audio cos: 0.000434 [Cossim] Audio cos: 0.004172
[Cossim] WAV stft_cos: 0.129544 samples: 120960 [Cossim] WAV stft_cos: 0.114254 samples: 120960
tests/clone-CUDA0-Q8_0.log
+19 -19
View File
@@ -69,13 +69,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] tts-pad-embed: [2048] first4: -0.005270 0.008716 -0.004355 0.001036 [Debug] tts-pad-embed: [2048] first4: -0.005270 0.008716 -0.004355 0.001036
[Debug] speaker-emb: [2048] first4: 0.098318 -0.023761 -0.011860 0.006760 [Debug] speaker-emb: [2048] first4: 0.098318 -0.023761 -0.011860 0.006760
[Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000 [Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000
[Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.467419 -0.069494 0.014718 -0.037177 [Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.467628 -0.069474 0.014061 -0.036574
[Debug] talker-hidden-prefill-l7: [225, 2048] first4: -2.191055 0.153858 0.210238 1.590774 [Debug] talker-hidden-prefill-l7: [225, 2048] first4: -2.188927 0.157751 0.206575 1.589241
[Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.963038 0.342889 0.321313 1.565405 [Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.961075 0.345972 0.318183 1.565162
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -1.446324 0.303191 0.670301 0.972597 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -1.449744 0.302516 0.671302 0.977163
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 6.668487 -34.319931 -7.307170 28.604122 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 6.664758 -34.346836 -7.308812 28.629059
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.226760 -1.218680 -0.230887 0.886599 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.226849 -1.220798 -0.231159 0.888217
[Debug] talker-logits-prefill: [3072] first4: 1.578662 -8.053411 -7.232630 -10.802841 [Debug] talker-logits-prefill: [3072] first4: 1.720009 -8.054232 -7.196761 -10.631991
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
@@ -94,7 +94,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=516 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=516 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.021538 -0.036059 0.003081 0.073169 [Debug] next-emb-step0: [2048] first4: 0.021538 -0.036059 0.003081 0.073169
[Debug] talker-hidden-step1: [2048] first4: 0.405619 2.517920 0.041025 0.037156 [Debug] talker-hidden-step1: [2048] first4: 0.475105 2.473964 -0.016539 -0.002217
[Sample] step=1 c0=1902 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.968624 max: 1.6806e-01 mean: 9.6580e-03 [Cossim] Embed cos: 0.968624 max: 1.6806e-01 mean: 9.6580e-03
[Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] L0 cos: 0.984967 max: 7.6842e-01 mean: 1.2814e-02 [Cossim] L0 cos: 0.984967 max: 7.6794e-01 mean: 1.2815e-02
[Cossim] L7 cos: 0.999978 max: 5.0762e+00 mean: 3.8277e-02 [Cossim] L7 cos: 0.999978 max: 3.8428e+00 mean: 3.8273e-02
[Cossim] L14 cos: 0.999955 max: 5.0315e+00 mean: 7.4583e-02 [Cossim] L14 cos: 0.999955 max: 3.8003e+00 mean: 7.4610e-02
[Cossim] L21 cos: 0.999567 max: 1.7787e+01 mean: 2.7419e-01 [Cossim] L21 cos: 0.999571 max: 1.7731e+01 mean: 2.7409e-01
[Cossim] L27 cos: 0.998136 max: 1.1157e+02 mean: 6.4898e-01 [Cossim] L27 cos: 0.998175 max: 1.0911e+02 mean: 6.4683e-01
[Cossim] Final cos: 0.991474 max: 1.6478e+01 mean: 2.0378e-01 [Cossim] Final cos: 0.991642 max: 1.5624e+01 mean: 2.0335e-01
[Cossim] Logits cos: 0.998908 max: 1.2573e+00 mean: 1.9487e-01 [Cossim] Logits cos: 0.999025 max: 1.2140e+00 mean: 1.8196e-01
[Cossim] NextEmbStep0 cos: 0.848334 max: 2.7271e-01 mean: 2.8037e-02 [Cossim] NextEmbStep0 cos: 0.848334 max: 2.7271e-01 mean: 2.8037e-02
[Cossim] TalkerHiddenStep1 cos: 0.992927 max: 1.6351e+00 mean: 2.5651e-01 [Cossim] TalkerHiddenStep1 cos: 0.993046 max: 1.5003e+00 mean: 2.5412e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 3.8069e-04 mean: 1.3207e-06 [Cossim] SpkMFA cos: 1.000000 max: 3.8069e-04 mean: 1.3207e-06
[Cossim] SpkASP cos: 1.000000 max: 7.8678e-06 mean: 2.2438e-07 [Cossim] SpkASP cos: 1.000000 max: 7.8678e-06 mean: 2.2438e-07
[Cossim] SpeakerEmb cos: 1.000000 max: 1.4305e-05 mean: 2.5522e-07 [Cossim] SpeakerEmb cos: 1.000000 max: 1.4305e-05 mean: 2.5522e-07
[Cossim] CodesFull exact: 4.07% (1008 values) [Cossim] CodesFull exact: 2.18% (1008 values)
[Cossim] Audio cos: 0.005833 [Cossim] Audio cos: 0.024229
[Cossim] WAV stft_cos: 0.209095 samples: 120960 [Cossim] WAV stft_cos: 0.230993 samples: 120960
tests/clone-Vulkan0-BF16.log
+53 -53
View File
@@ -70,47 +70,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] speaker-emb: [2048] first4: 0.098348 -0.023729 -0.011668 0.006747 [Debug] speaker-emb: [2048] first4: 0.098348 -0.023729 -0.011668 0.006747
[Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000 [Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000
[Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.461532 -0.069882 0.015988 -0.039215 [Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.461532 -0.069882 0.015988 -0.039215
[Debug] talker-hidden-prefill-l7: [225, 2048] first4: -1.523297 -0.014717 0.347053 1.776380 [Debug] talker-hidden-prefill-l7: [225, 2048] first4: -1.523296 -0.014714 0.347054 1.776378
[Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.294979 0.182117 0.448301 1.743050 [Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.294969 0.182105 0.448259 1.743060
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.769804 0.163546 0.799334 1.111618 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.769753 0.163422 0.799222 1.111621
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.050038 -34.400974 -7.071052 28.632160 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.050603 -34.401546 -7.070674 28.629971
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.239952 -1.222665 -0.223629 0.888272 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.239971 -1.222684 -0.223617 0.888203
[Debug] talker-logits-prefill: [3072] first4: 0.900955 -7.879465 -6.792802 -10.430634 [Debug] talker-logits-prefill: [3072] first4: 0.864874 -7.873147 -6.799634 -10.410440
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1203 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=203 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=400 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=447 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=240 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=2010 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=580 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=196 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=905 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1984 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=432 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=816 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1962 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1625 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=346 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=288 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=93 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=501 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1546 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=681 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=936 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=166 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=2002 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1523 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=906 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=121 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1203.000000 203.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.056839 -0.027733 0.030860 -0.040129 [Debug] next-emb-step0: [2048] first4: 0.004129 -0.009911 0.095512 0.080624
[Debug] talker-hidden-step1: [2048] first4: -0.632528 3.517379 -1.484728 0.181401 [Debug] talker-hidden-step1: [2048] first4: 0.140091 2.670658 -0.323406 0.280319
[Sample] step=1 c0=1570 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=682 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=814 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=367 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1906 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=1210 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=985 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1640 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=91 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=329 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1573 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=564 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=162 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1004 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=765 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1767 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=437 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=855 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=729 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1456 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=344 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1185 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=85 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=108 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=1435 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -120,8 +120,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1203.000000 203.000000 [Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: -0.001460 -0.002505 -0.002506 -0.001414 [Debug] output-audio: [122880] first4: -0.001118 -0.001998 -0.001883 -0.000914
[WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.962709 max: 1.9437e-01 mean: 1.0686e-02 [Cossim] Embed cos: 0.962709 max: 1.9437e-01 mean: 1.0686e-02
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.983223 max: 7.7067e-01 mean: 1.3991e-02 [Cossim] L0 cos: 0.983221 max: 7.7031e-01 mean: 1.3992e-02
[Cossim] L7 cos: 0.999989 max: 1.2859e+01 mean: 3.3234e-02 [Cossim] L7 cos: 0.999989 max: 1.2887e+01 mean: 3.3226e-02
[Cossim] L14 cos: 0.999964 max: 1.2860e+01 mean: 7.1854e-02 [Cossim] L14 cos: 0.999964 max: 1.2890e+01 mean: 7.1825e-02
[Cossim] L21 cos: 0.999532 max: 1.6955e+01 mean: 2.8042e-01 [Cossim] L21 cos: 0.999530 max: 1.6974e+01 mean: 2.8088e-01
[Cossim] L27 cos: 0.998217 max: 4.0720e+01 mean: 6.4082e-01 [Cossim] L27 cos: 0.998207 max: 4.0847e+01 mean: 6.4201e-01
[Cossim] Final cos: 0.990703 max: 1.5827e+01 mean: 2.0875e-01 [Cossim] Final cos: 0.990643 max: 1.5858e+01 mean: 2.0918e-01
[Cossim] Logits cos: 0.999316 max: 1.1188e+00 mean: 1.3424e-01 [Cossim] Logits cos: 0.999282 max: 1.1296e+00 mean: 1.3534e-01
[Cossim] NextEmbStep0 cos: 0.541941 max: 2.4936e-01 mean: 5.0259e-02 [Cossim] NextEmbStep0 cos: 0.820819 max: 1.6791e-01 mean: 2.9876e-02
[Cossim] TalkerHiddenStep1 cos: 0.520135 max: 2.0678e+01 mean: 1.9063e+00 [Cossim] TalkerHiddenStep1 cos: 0.992235 max: 2.2467e+00 mean: 2.6372e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 0.999979 max: 7.0797e-02 mean: 1.0370e-03 [Cossim] SpkMFA cos: 0.999979 max: 7.0797e-02 mean: 1.0370e-03
[Cossim] SpkASP cos: 1.000000 max: 2.3381e-03 mean: 1.5046e-04 [Cossim] SpkASP cos: 1.000000 max: 2.3381e-03 mean: 1.5046e-04
[Cossim] SpeakerEmb cos: 1.000000 max: 6.5649e-04 mean: 1.2850e-04 [Cossim] SpeakerEmb cos: 1.000000 max: 6.5649e-04 mean: 1.2850e-04
[Cossim] CodesFull exact: 0.30% (1008 values) [Cossim] CodesFull exact: 2.18% (1008 values)
[Cossim] Audio cos: -0.005066 [Cossim] Audio cos: -0.000684
[Cossim] WAV stft_cos: 0.148098 samples: 120960 [Cossim] WAV stft_cos: 0.197572 samples: 120960
tests/clone-Vulkan0-F32.log
+53 -53
View File
@@ -70,47 +70,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] speaker-emb: [2048] first4: 0.098351 -0.023771 -0.011823 0.006760 [Debug] speaker-emb: [2048] first4: 0.098351 -0.023771 -0.011823 0.006760
[Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000 [Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000
[Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.459179 -0.069735 0.016071 -0.038750 [Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.459179 -0.069735 0.016071 -0.038750
[Debug] talker-hidden-prefill-l7: [225, 2048] first4: -1.522404 -0.010386 0.338326 1.771031 [Debug] talker-hidden-prefill-l7: [225, 2048] first4: -1.522395 -0.010395 0.338314 1.771020
[Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.293831 0.185947 0.439517 1.737659 [Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.293835 0.186095 0.439391 1.737684
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.770112 0.166327 0.790798 1.105181 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -0.769963 0.166535 0.791349 1.104901
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.085222 -34.410065 -7.070534 28.603781 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 7.085280 -34.408360 -7.072290 28.600613
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.241214 -1.223315 -0.223672 0.887629 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.241213 -1.223239 -0.223725 0.887520
[Debug] talker-logits-prefill: [3072] first4: 0.986328 -7.724609 -6.861816 -10.246094 [Debug] talker-logits-prefill: [3072] first4: 0.996094 -7.732422 -6.857910 -10.236328
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1203 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=203 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=400 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=447 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=550 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1506 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=580 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=915 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=905 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=470 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=432 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1713 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1962 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1221 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=346 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=223 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=93 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1502 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1546 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1702 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=936 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=1860 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=2002 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1008 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=389 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=127 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1203.000000 203.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.066254 -0.031121 0.037696 -0.036131 [Debug] next-emb-step0: [2048] first4: 0.034006 -0.013787 0.039109 0.091488
[Debug] talker-hidden-step1: [2048] first4: -0.450483 3.398298 -1.360764 0.616965 [Debug] talker-hidden-step1: [2048] first4: 0.742317 2.656916 -0.143988 -0.179221
[Sample] step=1 c0=1570 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=441 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=814 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=318 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1630 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=350 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=980 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=985 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=672 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=853 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=51 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=329 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1534 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=564 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1490 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=692 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=59 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1958 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=437 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1728 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=729 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1787 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=344 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=678 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=85 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=381 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=1668 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -120,8 +120,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1203.000000 203.000000 [Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: -0.001441 -0.002432 -0.002447 -0.001549 [Debug] output-audio: [122880] first4: -0.001355 -0.002290 -0.002015 -0.000875
[WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.972129 max: 1.9482e-01 mean: 8.1107e-03 [Cossim] Embed cos: 0.972129 max: 1.9482e-01 mean: 8.1107e-03
[Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TrailingText cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.0431e-06 mean: 9.3616e-09
[Cossim] L0 cos: 0.987066 max: 6.8535e-01 mean: 1.0819e-02 [Cossim] L0 cos: 0.987066 max: 6.8538e-01 mean: 1.0819e-02
[Cossim] L7 cos: 0.999991 max: 1.0628e+01 mean: 2.7183e-02 [Cossim] L7 cos: 0.999991 max: 1.4623e+01 mean: 2.7206e-02
[Cossim] L14 cos: 0.999972 max: 1.0626e+01 mean: 5.9813e-02 [Cossim] L14 cos: 0.999971 max: 1.4620e+01 mean: 5.9841e-02
[Cossim] L21 cos: 0.999708 max: 1.0632e+01 mean: 2.2441e-01 [Cossim] L21 cos: 0.999707 max: 1.4630e+01 mean: 2.2443e-01
[Cossim] L27 cos: 0.998904 max: 2.5903e+01 mean: 5.2200e-01 [Cossim] L27 cos: 0.998904 max: 2.6191e+01 mean: 5.2204e-01
[Cossim] Final cos: 0.994364 max: 7.8407e+00 mean: 1.6916e-01 [Cossim] Final cos: 0.994357 max: 7.9427e+00 mean: 1.6918e-01
[Cossim] Logits cos: 0.999523 max: 8.6606e-01 mean: 1.2265e-01 [Cossim] Logits cos: 0.999565 max: 8.3774e-01 mean: 1.1954e-01
[Cossim] NextEmbStep0 cos: 0.541281 max: 2.5191e-01 mean: 5.0209e-02 [Cossim] NextEmbStep0 cos: 0.822712 max: 2.7881e-01 mean: 3.0097e-02
[Cossim] TalkerHiddenStep1 cos: 0.519919 max: 2.0566e+01 mean: 1.9084e+00 [Cossim] TalkerHiddenStep1 cos: 0.990734 max: 1.7581e+00 mean: 2.9635e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 8.9911e-03 mean: 1.3077e-04 [Cossim] SpkMFA cos: 1.000000 max: 8.9911e-03 mean: 1.3077e-04
[Cossim] SpkASP cos: 1.000000 max: 2.2933e-04 mean: 1.7414e-05 [Cossim] SpkASP cos: 1.000000 max: 2.2933e-04 mean: 1.7414e-05
[Cossim] SpeakerEmb cos: 1.000000 max: 7.9088e-05 mean: 1.3439e-05 [Cossim] SpeakerEmb cos: 1.000000 max: 7.9088e-05 mean: 1.3439e-05
[Cossim] CodesFull exact: 0.20% (1008 values) [Cossim] CodesFull exact: 3.08% (1008 values)
[Cossim] Audio cos: -0.005021 [Cossim] Audio cos: 0.012894
[Cossim] WAV stft_cos: 0.156799 samples: 120960 [Cossim] WAV stft_cos: 0.219480 samples: 120960
tests/clone-Vulkan0-Q4_K_M.log
+53 -53
View File
@@ -70,47 +70,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] speaker-emb: [2048] first4: 0.098351 -0.023771 -0.011823 0.006760 [Debug] speaker-emb: [2048] first4: 0.098351 -0.023771 -0.011823 0.006760
[Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 696.000000 1522.000000 [Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 696.000000 1522.000000
[Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.437219 -0.067416 0.007625 0.014900 [Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.437219 -0.067416 0.007625 0.014900
[Debug] talker-hidden-prefill-l7: [225, 2048] first4: -8.841837 0.789566 0.499846 0.089505 [Debug] talker-hidden-prefill-l7: [225, 2048] first4: -8.840106 0.790002 0.499952 0.090613
[Debug] talker-hidden-prefill-l14: [225, 2048] first4: -8.777140 1.123894 0.603008 0.155361 [Debug] talker-hidden-prefill-l14: [225, 2048] first4: -8.775373 1.124294 0.603039 0.156350
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -8.660569 1.697090 1.104082 -0.795212 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -8.657483 1.697616 1.103059 -0.794886
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 4.911671 -31.767857 -6.812796 21.723780 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 4.921110 -31.765329 -6.813909 21.710571
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.165718 -1.119261 -0.213588 0.668089 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.166035 -1.119164 -0.213621 0.667677
[Debug] talker-logits-prefill: [3072] first4: -0.156006 -5.531250 -7.037109 -10.544922 [Debug] talker-logits-prefill: [3072] first4: -0.148926 -5.533691 -7.051758 -10.552734
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1203 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=647 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=417 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1674 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=442 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1412 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1674 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=823 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=75 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=470 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1763 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=996 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=936 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=318 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=819 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=223 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=761 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=426 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=755 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=625 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=79 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=866 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=37 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1456 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=1789 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=1047 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1203.000000 647.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 417.000000
[Debug] next-emb-step0: [2048] first4: -0.013052 -0.001298 0.002898 -0.014310 [Debug] next-emb-step0: [2048] first4: 0.005741 0.034089 -0.000187 0.073664
[Debug] talker-hidden-step1: [2048] first4: -0.326902 2.695708 -1.438280 0.962824 [Debug] talker-hidden-step1: [2048] first4: 0.574148 2.664108 0.812643 0.832140
[Sample] step=1 c0=1570 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=682 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1094 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=203 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1226 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=405 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=342 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=390 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1487 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=447 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=131 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1059 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=638 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=564 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1573 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1004 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=506 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=282 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=1887 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1603 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1713 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=122 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=338 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1947 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=715 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=1605 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=1842 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -120,8 +120,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1203.000000 647.000000 [Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 417.000000
[Debug] output-audio: [122880] first4: -0.000863 -0.001503 -0.001410 -0.000697 [Debug] output-audio: [122880] first4: -0.001181 -0.002022 -0.001853 -0.000890
[WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.917129 max: 2.2247e-01 mean: 2.1853e-02 [Cossim] Embed cos: 0.917129 max: 2.2247e-01 mean: 2.1853e-02
[Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TrailingText cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04 [Cossim] TTSPadEmbed cos: 0.999969 max: 1.4858e-03 mean: 2.2947e-04
[Cossim] L0 cos: 0.961477 max: 8.5747e-01 mean: 2.8366e-02 [Cossim] L0 cos: 0.961477 max: 8.5893e-01 mean: 2.8365e-02
[Cossim] L7 cos: 0.997433 max: 3.9387e+01 mean: 1.6338e-01 [Cossim] L7 cos: 0.997433 max: 3.9393e+01 mean: 1.6338e-01
[Cossim] L14 cos: 0.997359 max: 3.8752e+01 mean: 2.4737e-01 [Cossim] L14 cos: 0.997359 max: 3.8757e+01 mean: 2.4740e-01
[Cossim] L21 cos: 0.995878 max: 3.9000e+01 mean: 7.7237e-01 [Cossim] L21 cos: 0.995876 max: 3.9001e+01 mean: 7.7270e-01
[Cossim] L27 cos: 0.956754 max: 1.7476e+03 mean: 1.9744e+00 [Cossim] L27 cos: 0.956681 max: 1.7493e+03 mean: 1.9751e+00
[Cossim] Final cos: 0.955802 max: 3.3789e+01 mean: 5.4922e-01 [Cossim] Final cos: 0.955770 max: 3.3637e+01 mean: 5.4943e-01
[Cossim] Logits cos: 0.992553 max: 2.9861e+00 mean: 4.5271e-01 [Cossim] Logits cos: 0.992590 max: 2.9793e+00 mean: 4.5130e-01
[Cossim] NextEmbStep0 cos: 0.582183 max: 2.0788e-01 mean: 4.7545e-02 [Cossim] NextEmbStep0 cos: 0.819102 max: 1.4990e-01 mean: 3.0364e-02
[Cossim] TalkerHiddenStep1 cos: 0.505559 max: 2.3324e+01 mean: 1.9119e+00 [Cossim] TalkerHiddenStep1 cos: 0.957365 max: 3.9930e+00 mean: 6.2978e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 8.9911e-03 mean: 1.3077e-04 [Cossim] SpkMFA cos: 1.000000 max: 8.9911e-03 mean: 1.3077e-04
[Cossim] SpkASP cos: 1.000000 max: 2.2933e-04 mean: 1.7414e-05 [Cossim] SpkASP cos: 1.000000 max: 2.2933e-04 mean: 1.7414e-05
[Cossim] SpeakerEmb cos: 1.000000 max: 7.9088e-05 mean: 1.3439e-05 [Cossim] SpeakerEmb cos: 1.000000 max: 7.9088e-05 mean: 1.3439e-05
[Cossim] CodesFull exact: 0.30% (1008 values) [Cossim] CodesFull exact: 1.79% (1008 values)
[Cossim] Audio cos: -0.000113 [Cossim] Audio cos: -0.007409
[Cossim] WAV stft_cos: 0.035981 samples: 120960 [Cossim] WAV stft_cos: 0.180735 samples: 120960
tests/clone-Vulkan0-Q8_0.log
+53 -53
View File
@@ -70,47 +70,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] speaker-emb: [2048] first4: 0.098351 -0.023771 -0.011823 0.006760 [Debug] speaker-emb: [2048] first4: 0.098351 -0.023771 -0.011823 0.006760
[Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000 [Debug] ref-codes: [16, 215] first4: 1221.000000 2042.000000 509.000000 1522.000000
[Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.463271 -0.072727 0.015013 -0.037542 [Debug] talker-hidden-prefill-l0: [225, 2048] first4: -0.463271 -0.072727 0.015013 -0.037542
[Debug] talker-hidden-prefill-l7: [225, 2048] first4: -2.203949 0.171320 0.187689 1.621109 [Debug] talker-hidden-prefill-l7: [225, 2048] first4: -2.203950 0.171320 0.187689 1.621109
[Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.980078 0.357584 0.294047 1.595288 [Debug] talker-hidden-prefill-l14: [225, 2048] first4: -1.980177 0.357490 0.294066 1.595268
[Debug] talker-hidden-prefill-l21: [225, 2048] first4: -1.466878 0.293497 0.651744 1.015960 [Debug] talker-hidden-prefill-l21: [225, 2048] first4: -1.466965 0.293282 0.651436 1.016212
[Debug] talker-hidden-prefill-l27: [225, 2048] first4: 6.708439 -34.419636 -7.427876 28.535091 [Debug] talker-hidden-prefill-l27: [225, 2048] first4: 6.705817 -34.413361 -7.426254 28.536070
[Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.228206 -1.222688 -0.234791 0.884797 [Debug] talker-hidden-prefill-final: [225, 2048] first4: 0.228129 -1.222527 -0.234752 0.884873
[Debug] talker-logits-prefill: [3072] first4: 1.251465 -7.830078 -7.050781 -10.621094 [Debug] talker-logits-prefill: [3072] first4: 1.258301 -7.826172 -7.041016 -10.646484
[Sample] step=0 c0=27 u=-1.0000000000 subseq=0 [Sample] step=0 c0=27 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1408 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1203 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1824 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=203 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=400 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=447 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=550 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1506 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=580 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=1452 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=905 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=470 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=432 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1713 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1962 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1221 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=346 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=18 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=93 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1502 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1546 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=936 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=411 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=2002 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=906 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=516 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 27.000000 1408.000000 1203.000000 203.000000 [Debug] codes-step0: [16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] next-emb-step0: [2048] first4: 0.057144 -0.027413 0.030649 -0.040303 [Debug] next-emb-step0: [2048] first4: 0.021538 -0.036059 0.003081 0.073169
[Debug] talker-hidden-step1: [2048] first4: -0.610442 3.379674 -1.551872 0.230346 [Debug] talker-hidden-step1: [2048] first4: 0.373818 2.559147 -0.075776 0.173605
[Sample] step=1 c0=1570 u=-1.0000000000 subseq=16 [Sample] step=1 c0=1902 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=682 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=814 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=2025 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=367 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=324 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1906 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=350 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=985 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=356 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=91 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1544 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1573 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=162 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1004 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=765 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1702 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1767 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=437 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=855 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=729 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1306 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=781 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=2007 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=85 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=4 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=1435 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -120,8 +120,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1203.000000 203.000000 [Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: -0.001439 -0.002481 -0.002478 -0.001377 [Debug] output-audio: [122880] first4: -0.001815 -0.003137 -0.002853 -0.001444
[WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/clone/clone-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/clone/clone-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -145,15 +145,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.970174 max: 1.9000e-01 mean: 8.9735e-03 [Cossim] Embed cos: 0.970174 max: 1.9000e-01 mean: 8.9735e-03
[Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TrailingText cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 2.1233e-04 mean: 4.3863e-05
[Cossim] L0 cos: 0.985828 max: 7.7037e-01 mean: 1.1813e-02 [Cossim] L0 cos: 0.985828 max: 7.7027e-01 mean: 1.1813e-02
[Cossim] L7 cos: 0.999978 max: 9.3408e+00 mean: 3.6074e-02 [Cossim] L7 cos: 0.999978 max: 9.3564e+00 mean: 3.6075e-02
[Cossim] L14 cos: 0.999956 max: 9.3533e+00 mean: 7.0647e-02 [Cossim] L14 cos: 0.999956 max: 9.3684e+00 mean: 7.0631e-02
[Cossim] L21 cos: 0.999606 max: 1.8653e+01 mean: 2.5031e-01 [Cossim] L21 cos: 0.999605 max: 1.8669e+01 mean: 2.5064e-01
[Cossim] L27 cos: 0.998551 max: 4.8180e+01 mean: 5.6835e-01 [Cossim] L27 cos: 0.998549 max: 4.7575e+01 mean: 5.6932e-01
[Cossim] Final cos: 0.992873 max: 1.9007e+01 mean: 1.7984e-01 [Cossim] Final cos: 0.992851 max: 1.9025e+01 mean: 1.8017e-01
[Cossim] Logits cos: 0.999578 max: 7.9731e-01 mean: 1.0601e-01 [Cossim] Logits cos: 0.999577 max: 8.0415e-01 mean: 1.0617e-01
[Cossim] NextEmbStep0 cos: 0.541937 max: 2.4933e-01 mean: 5.0250e-02 [Cossim] NextEmbStep0 cos: 0.848334 max: 2.7271e-01 mean: 2.8037e-02
[Cossim] TalkerHiddenStep1 cos: 0.520921 max: 2.0466e+01 mean: 1.9067e+00 [Cossim] TalkerHiddenStep1 cos: 0.993818 max: 1.5023e+00 mean: 2.3923e-01
[Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08 [Cossim] MelHann cos: 1.000000 max: 1.7881e-07 mean: 3.2043e-08
[Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10 [Cossim] MelBasis cos: 1.000000 max: 1.7881e-07 mean: 3.9379e-10
[Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07 [Cossim] MelMag cos: 1.000000 max: 3.1307e-05 mean: 5.6396e-07
@@ -171,6 +171,6 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] SpkMFA cos: 1.000000 max: 8.9911e-03 mean: 1.3077e-04 [Cossim] SpkMFA cos: 1.000000 max: 8.9911e-03 mean: 1.3077e-04
[Cossim] SpkASP cos: 1.000000 max: 2.2933e-04 mean: 1.7414e-05 [Cossim] SpkASP cos: 1.000000 max: 2.2933e-04 mean: 1.7414e-05
[Cossim] SpeakerEmb cos: 1.000000 max: 7.9088e-05 mean: 1.3439e-05 [Cossim] SpeakerEmb cos: 1.000000 max: 7.9088e-05 mean: 1.3439e-05
[Cossim] CodesFull exact: 0.20% (1008 values) [Cossim] CodesFull exact: 3.97% (1008 values)
[Cossim] Audio cos: 0.003500 [Cossim] Audio cos: 0.003665
[Cossim] WAV stft_cos: 0.135950 samples: 120960 [Cossim] WAV stft_cos: 0.224321 samples: 120960
tests/customvoice-CPU-BF16.log
+124
View File
@@ -0,0 +1,124 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-customvoice-BF16.gguf: 404 tensors, data at offset 5949824
[WeightCtx] Loaded 316 tensors, 3322.0 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 334.1 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-BF16.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=custom_voice tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=9
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=41, hidden=2048, lang=english (id=2050), speaker=vivian (id=3065) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.383354 -0.060806 0.002336 -0.013838
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.343848 0.256815 1.585862 2.190261
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.103534 0.419664 1.685469 2.136022
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.675360 0.328115 2.041473 1.583337
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.927499 -24.485857 -2.206259 26.156494
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.028593 -0.788262 -0.063200 0.735004
[Debug] talker-logits-prefill: [3072] first4: -0.728725 -8.061918 -2.771915 -5.009385
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.039671 -0.032533 -0.000920 0.180072
[Debug] talker-hidden-step1: [2048] first4: 0.535159 -1.850773 3.023841 -1.191638
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1191 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=667 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=943 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1177 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1907 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=944 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1711 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=99 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1070 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=194 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000016 0.000020 0.000017 0.000019
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Speaker: vivian
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-customvoice-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/customvoice/customvoice-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08
[Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] L0 cos: 1.000000 max: 4.9011e-02 mean: 1.1031e-04
[Cossim] L7 cos: 0.999997 max: 2.0044e+01 mean: 1.9537e-03
[Cossim] L14 cos: 0.999997 max: 2.0063e+01 mean: 3.7767e-03
[Cossim] L21 cos: 0.999997 max: 2.0063e+01 mean: 1.1207e-02
[Cossim] L27 cos: 0.999995 max: 1.9309e+01 mean: 6.0608e-02
[Cossim] Final cos: 0.999991 max: 4.9782e-01 mean: 4.1352e-03
[Cossim] Logits cos: 0.999998 max: 4.0338e-02 mean: 6.8873e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.3709e-06 mean: 9.5057e-09
[Cossim] TalkerHiddenStep1 cos: 0.999994 max: 3.4467e-02 mean: 6.2675e-03
[Cossim] CodesFull exact: 18.85% (1008 values)
[Cossim] Audio cos: 0.175083
[Cossim] WAV stft_cos: 0.437989 samples: 120960
tests/customvoice-CPU-F32.log
+124
View File
@@ -0,0 +1,124 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-customvoice-F32.gguf: 404 tensors, data at offset 5949728
[WeightCtx] Loaded 316 tensors, 6643.5 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 668.1 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-F32.gguf: 398 tensors, data at offset 29664
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 116.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=custom_voice tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=9
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=41, hidden=2048, lang=english (id=2050), speaker=vivian (id=3065) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.383106 -0.060893 0.002297 -0.013863
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.343932 0.253800 1.586739 2.194738
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.103490 0.416969 1.686401 2.140801
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.674539 0.324556 2.041308 1.586615
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.924737 -24.438339 -2.204346 26.094110
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.028519 -0.787023 -0.063169 0.733522
[Debug] talker-logits-prefill: [3072] first4: -0.712858 -8.073822 -2.787058 -5.005693
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.039671 -0.032533 -0.000920 0.180072
[Debug] talker-hidden-step1: [2048] first4: 0.528140 -1.855237 3.012427 -1.174796
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1191 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=667 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=943 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1177 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1907 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=944 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1711 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=99 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1070 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=194 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000014 0.000016
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Speaker: vivian
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav
[Quant] F32 -> ../models/qwen-talker-1.7b-customvoice-F32.gguf + ../models/qwen-tokenizer-12hz-F32.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-F32.gguf --codec ../models/qwen-tokenizer-12hz-F32.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/customvoice/customvoice-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08
[Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] L0 cos: 1.000000 max: 2.2888e-05 mean: 7.4999e-08
[Cossim] L7 cos: 1.000000 max: 2.4414e-03 mean: 5.1331e-07
[Cossim] L14 cos: 1.000000 max: 2.4414e-03 mean: 1.0756e-06
[Cossim] L21 cos: 1.000000 max: 2.4414e-03 mean: 3.6395e-06
[Cossim] L27 cos: 1.000000 max: 4.2725e-03 mean: 1.7006e-05
[Cossim] Final cos: 1.000000 max: 5.9128e-05 mean: 1.2670e-06
[Cossim] Logits cos: 1.000000 max: 1.6212e-05 mean: 2.7734e-06
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.3709e-06 mean: 9.5057e-09
[Cossim] TalkerHiddenStep1 cos: 1.000000 max: 1.5259e-05 mean: 2.4196e-06
[Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 1.000000
[Cossim] WAV stft_cos: 1.000000 samples: 120960
tests/customvoice-CPU-Q4_K_M.log
+124
View File
@@ -0,0 +1,124 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-customvoice-Q4_K_M.gguf: 404 tensors, data at offset 5949824
[WeightCtx] Loaded 316 tensors, 1012.7 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 109.5 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-Q4_K_M.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 16.4 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 25.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 13.6 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=custom_voice tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=9
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=41, hidden=2048, lang=english (id=2050), speaker=vivian (id=3065) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [41, 2048] first4: 0.020406 -0.009921 0.006129 -0.021787
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003884 0.008386 -0.004502 -0.000786
[Debug] tts-pad-embed: [2048] first4: -0.003884 0.008386 -0.004502 -0.000786
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.350445 -0.044229 -0.032651 0.005133
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: 8.604717 -0.493292 6.315798 14.766369
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: 8.589753 -0.216029 6.477687 14.595556
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: 8.620410 0.116378 6.737935 14.118506
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: 10.463799 -29.392971 -0.921735 38.965904
[Debug] talker-hidden-prefill-final: [41, 2048] first4: 0.321839 -0.944053 -0.026343 1.092428
[Debug] talker-logits-prefill: [3072] first4: -0.966614 -8.205486 -1.614648 -5.346405
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.041045 -0.031131 0.000245 0.179301
[Debug] talker-hidden-step1: [2048] first4: 1.993904 -1.744391 3.053013 -1.318921
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=821 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1631 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=859 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1138 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=122 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1842 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1252 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=1532 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=501 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1567 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1060 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1799 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=570 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000013 0.000016
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Speaker: vivian
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav
[Quant] Q4_K_M -> ../models/qwen-talker-1.7b-customvoice-Q4_K_M.gguf + ../models/qwen-tokenizer-12hz-Q4_K_M.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-Q4_K_M.gguf --codec ../models/qwen-tokenizer-12hz-Q4_K_M.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/customvoice/customvoice-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 0.999698 max: 1.6597e-01 mean: 1.0155e-03
[Cossim] TrailingText cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04
[Cossim] TTSPadEmbed cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04
[Cossim] L0 cos: 0.999236 max: 7.3277e-01 mean: 6.4444e-03
[Cossim] L7 cos: 0.997571 max: 9.0228e+01 mean: 5.9683e-01
[Cossim] L14 cos: 0.997552 max: 9.0203e+01 mean: 7.1903e-01
[Cossim] L21 cos: 0.997045 max: 9.4607e+01 mean: 1.3416e+00
[Cossim] L27 cos: 0.981897 max: 8.2715e+02 mean: 5.7338e+00
[Cossim] Final cos: 0.903759 max: 6.4361e+01 mean: 4.1779e-01
[Cossim] Logits cos: 0.989976 max: 3.0350e+00 mean: 4.6430e-01
[Cossim] NextEmbStep0 cos: 0.999915 max: 3.2489e-03 mean: 7.7490e-04
[Cossim] TalkerHiddenStep1 cos: 0.962386 max: 2.2756e+00 mean: 4.8091e-01
[Cossim] CodesFull exact: 3.17% (1008 values)
[Cossim] Audio cos: -0.032617
[Cossim] WAV stft_cos: 0.098593 samples: 120960
tests/customvoice-CPU-Q8_0.log
+124
View File
@@ -0,0 +1,124 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-customvoice-Q8_0.gguf: 404 tensors, data at offset 5949824
[WeightCtx] Loaded 316 tensors, 1765.0 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 177.5 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-Q8_0.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 30.9 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 33.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 25.6 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=custom_voice tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=9
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=41, hidden=2048, lang=english (id=2050), speaker=vivian (id=3065) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [41, 2048] first4: 0.021601 -0.009355 0.007161 -0.019744
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003786 0.008339 -0.005023 -0.000287
[Debug] tts-pad-embed: [2048] first4: -0.003786 0.008339 -0.005023 -0.000287
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.385550 -0.059227 0.003153 -0.015722
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -2.096292 0.136028 2.073626 1.093323
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.841977 0.296974 2.177680 1.048685
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -1.422673 0.233760 2.557228 0.556667
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -1.257986 -24.830193 -1.723763 25.083599
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.038877 -0.801303 -0.049500 0.706581
[Debug] talker-logits-prefill: [3072] first4: -0.937246 -7.968508 -2.695894 -4.725146
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.039680 -0.032027 -0.001058 0.180065
[Debug] talker-hidden-step1: [2048] first4: 0.458236 -1.977294 3.021161 -1.306485
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1191 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=667 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=943 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1177 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1907 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=944 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1711 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=99 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1070 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=194 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000015 0.000017
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Speaker: vivian
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav
[Quant] Q8_0 -> ../models/qwen-talker-1.7b-customvoice-Q8_0.gguf + ../models/qwen-tokenizer-12hz-Q8_0.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-Q8_0.gguf --codec ../models/qwen-tokenizer-12hz-Q8_0.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/customvoice/customvoice-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 0.999962 max: 5.1095e-02 mean: 2.0739e-04
[Cossim] TrailingText cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05
[Cossim] L0 cos: 0.999985 max: 5.0579e-02 mean: 8.8635e-04
[Cossim] L7 cos: 0.999985 max: 1.0991e+01 mean: 4.6925e-02
[Cossim] L14 cos: 0.999985 max: 1.1005e+01 mean: 6.0868e-02
[Cossim] L21 cos: 0.999980 max: 1.1358e+01 mean: 1.2149e-01
[Cossim] L27 cos: 0.999866 max: 5.1759e+01 mean: 5.5204e-01
[Cossim] Final cos: 0.998199 max: 1.2530e+01 mean: 4.4798e-02
[Cossim] Logits cos: 0.999809 max: 3.7629e-01 mean: 7.2518e-02
[Cossim] NextEmbStep0 cos: 0.999992 max: 1.1319e-03 mean: 2.3562e-04
[Cossim] TalkerHiddenStep1 cos: 0.999182 max: 7.1951e-01 mean: 6.7702e-02
[Cossim] CodesFull exact: 5.06% (1008 values)
[Cossim] Audio cos: 0.080945
[Cossim] WAV stft_cos: 0.162718 samples: 120960
tests/customvoice-CUDA0-BF16.log
+96 -1
View File
@@ -16,7 +16,85 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512 [Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend [WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[CausalTransConv] FATAL: 'tok_dec.upsample.0.conv.weight' expected F32 or F16, got type 30 [WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=custom_voice tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=9
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=41, hidden=2048, lang=english (id=2050), speaker=vivian (id=3065) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.379817 -0.060882 0.002518 -0.014104
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.347125 0.265610 1.567530 2.175550
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.106186 0.429259 1.666740 2.121611
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.676503 0.336397 2.023456 1.568729
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.892239 -24.659126 -2.221928 26.377346
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.027560 -0.795376 -0.063772 0.742645
[Debug] talker-logits-prefill: [3072] first4: -0.707031 -8.062500 -2.812500 -4.968750
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.039671 -0.032533 -0.000920 0.180072
[Debug] talker-hidden-step1: [2048] first4: 0.504843 -1.868504 3.021610 -1.168490
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1191 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=667 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=943 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1177 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1907 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=944 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1711 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=99 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1070 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=194 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000016 0.000020 0.000016 0.000017
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Speaker: vivian [Input] Speaker: vivian
[Input] Lang: english Seed: 42 MaxNewTokens: 64 [Input] Lang: english Seed: 42 MaxNewTokens: 64
@@ -27,3 +105,20 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav [Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-customvoice-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf [Quant] BF16 -> ../models/qwen-talker-1.7b-customvoice-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy [GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/customvoice/customvoice-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08
[Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] L0 cos: 0.999997 max: 1.4418e-01 mean: 2.5783e-04
[Cossim] L7 cos: 0.999997 max: 1.8768e+01 mean: 3.1115e-03
[Cossim] L14 cos: 0.999997 max: 1.8776e+01 mean: 6.2973e-03
[Cossim] L21 cos: 0.999997 max: 1.8805e+01 mean: 2.1139e-02
[Cossim] L27 cos: 0.999990 max: 3.0633e+01 mean: 1.0479e-01
[Cossim] Final cos: 0.999970 max: 5.9777e-01 mean: 7.8789e-03
[Cossim] Logits cos: 0.999990 max: 9.4216e-02 mean: 1.7135e-02
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.3709e-06 mean: 9.5057e-09
[Cossim] TalkerHiddenStep1 cos: 0.999989 max: 4.3977e-02 mean: 8.0388e-03
[Cossim] CodesFull exact: 30.65% (1008 values)
[Cossim] Audio cos: 0.286480
[Cossim] WAV stft_cos: 0.419330 samples: 120960
tests/customvoice-CUDA0-F32.log
+27 -27
View File
@@ -18,15 +18,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend [WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7 [Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 199.8 MB into backend [WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 199.8 MB [DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 6.0 MB into backend [WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 48.2 MB into backend [WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 48.2 MB [SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend [WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000 [EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 4.0 MB into backend [WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 4.0 MB [EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend [WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB [EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz [Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
@@ -40,13 +40,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719 [Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409 [Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409 [Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.383106 -0.060893 0.002297 -0.013863 [Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.383101 -0.060891 0.002305 -0.013872
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.343934 0.253801 1.586739 2.194738 [Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.343905 0.253641 1.586693 2.194888
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.103492 0.416971 1.686401 2.140801 [Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.103505 0.416739 1.686352 2.140942
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.674541 0.324558 2.041308 1.586615 [Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.674517 0.324479 2.041196 1.586640
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.924734 -24.438322 -2.204347 26.094097 [Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.925121 -24.436701 -2.204262 26.093637
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.028519 -0.787022 -0.063169 0.733522 [Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.028531 -0.786979 -0.063167 0.733517
[Debug] talker-logits-prefill: [3072] first4: -0.712860 -8.073814 -2.787057 -5.005689 [Debug] talker-logits-prefill: [3072] first4: -0.705178 -8.074849 -2.788742 -5.008213
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
@@ -65,7 +65,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.039671 -0.032533 -0.000920 0.180072 [Debug] next-emb-step0: [2048] first4: -0.039671 -0.032533 -0.000920 0.180072
[Debug] talker-hidden-step1: [2048] first4: 0.528143 -1.855237 3.012431 -1.174793 [Debug] talker-hidden-step1: [2048] first4: 0.533395 -1.854379 3.011211 -1.176704
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16 [Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
@@ -92,7 +92,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000015 0.000017 [Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000014 0.000016
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -110,15 +110,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08 [Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08
[Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09 [Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] L0 cos: 1.000000 max: 3.0518e-05 mean: 7.9042e-08 [Cossim] L0 cos: 1.000000 max: 1.1292e-03 mean: 2.0020e-05
[Cossim] L7 cos: 1.000000 max: 4.8828e-04 mean: 4.0060e-07 [Cossim] L7 cos: 1.000000 max: 1.4209e-01 mean: 1.6547e-04
[Cossim] L14 cos: 1.000000 max: 4.8828e-04 mean: 1.0316e-06 [Cossim] L14 cos: 1.000000 max: 1.4233e-01 mean: 4.6968e-04
[Cossim] L21 cos: 1.000000 max: 4.8828e-04 mean: 3.9015e-06 [Cossim] L21 cos: 1.000000 max: 1.4513e-01 mean: 1.6701e-03
[Cossim] L27 cos: 1.000000 max: 3.9062e-03 mean: 1.6371e-05 [Cossim] L27 cos: 1.000000 max: 3.2617e+00 mean: 6.8250e-03
[Cossim] Final cos: 1.000000 max: 1.7715e-04 mean: 1.2082e-06 [Cossim] Final cos: 1.000000 max: 2.8135e-02 mean: 5.1405e-04
[Cossim] Logits cos: 1.000000 max: 1.2875e-05 mean: 2.3994e-06 [Cossim] Logits cos: 1.000000 max: 1.2022e-02 mean: 2.0501e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.3709e-06 mean: 9.5057e-09 [Cossim] NextEmbStep0 cos: 1.000000 max: 1.3709e-06 mean: 9.5057e-09
[Cossim] TalkerHiddenStep1 cos: 1.000000 max: 3.0518e-05 mean: 2.1401e-06 [Cossim] TalkerHiddenStep1 cos: 0.999999 max: 1.5196e-02 mean: 2.0715e-03
[Cossim] CodesFull exact: 100.00% (1008 values) [Cossim] CodesFull exact: 58.63% (1008 values)
[Cossim] Audio cos: 0.999999 [Cossim] Audio cos: 0.564482
[Cossim] WAV stft_cos: 1.000000 samples: 120960 [Cossim] WAV stft_cos: 0.705637 samples: 120960
tests/customvoice-CUDA0-Q4_K_M.log
+30 -30
View File
@@ -40,13 +40,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [41, 2048] first4: 0.020406 -0.009921 0.006129 -0.021787 [Debug] talker-input-embed: [41, 2048] first4: 0.020406 -0.009921 0.006129 -0.021787
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003884 0.008386 -0.004502 -0.000786 [Debug] trailing-text-hidden: [1, 2048] first4: -0.003884 0.008386 -0.004502 -0.000786
[Debug] tts-pad-embed: [2048] first4: -0.003884 0.008386 -0.004502 -0.000786 [Debug] tts-pad-embed: [2048] first4: -0.003884 0.008386 -0.004502 -0.000786
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.346090 -0.051518 -0.040558 -0.001631 [Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.347301 -0.052077 -0.041220 -0.002471
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: 8.658912 -0.418502 6.296154 14.764099 [Debug] talker-hidden-prefill-l7: [41, 2048] first4: 8.485819 -0.522394 6.125673 14.590284
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: 8.646750 -0.138280 6.456070 14.578581 [Debug] talker-hidden-prefill-l14: [41, 2048] first4: 8.475321 -0.244497 6.287801 14.403727
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: 8.547407 0.151086 6.711766 14.174829 [Debug] talker-hidden-prefill-l21: [41, 2048] first4: 8.389411 0.036782 6.543781 13.993382
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: 10.425589 -28.933830 -0.986194 39.437996 [Debug] talker-hidden-prefill-l27: [41, 2048] first4: 10.127118 -29.085281 -1.204731 39.292091
[Debug] talker-hidden-prefill-final: [41, 2048] first4: 0.320198 -0.927956 -0.028144 1.104057 [Debug] talker-hidden-prefill-final: [41, 2048] first4: 0.311438 -0.934033 -0.034426 1.101411
[Debug] talker-logits-prefill: [3072] first4: -0.933804 -8.354171 -1.749861 -5.627195 [Debug] talker-logits-prefill: [3072] first4: -0.810304 -8.010923 -1.648036 -5.381549
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
@@ -65,22 +65,22 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.041045 -0.031131 0.000245 0.179301 [Debug] next-emb-step0: [2048] first4: -0.041045 -0.031131 0.000245 0.179301
[Debug] talker-hidden-step1: [2048] first4: 2.075958 -1.768331 3.046258 -1.300610 [Debug] talker-hidden-step1: [2048] first4: 1.952667 -1.897995 3.202719 -1.181960
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16 [Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=821 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=821 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=362 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=1631 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=867 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=859 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1042 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1138 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=43 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=122 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1617 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=1842 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=484 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1252 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=294 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1532 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=347 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=501 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=125 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1567 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=12 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1060 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1339 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=1799 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=570 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=570 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
@@ -110,15 +110,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999698 max: 1.6597e-01 mean: 1.0155e-03 [Cossim] Embed cos: 0.999698 max: 1.6597e-01 mean: 1.0155e-03
[Cossim] TrailingText cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04 [Cossim] TrailingText cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04
[Cossim] TTSPadEmbed cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04 [Cossim] TTSPadEmbed cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04
[Cossim] L0 cos: 0.999289 max: 4.0790e-01 mean: 6.3178e-03 [Cossim] L0 cos: 0.999288 max: 3.9823e-01 mean: 6.3234e-03
[Cossim] L7 cos: 0.997569 max: 7.2079e+01 mean: 5.9714e-01 [Cossim] L7 cos: 0.997565 max: 7.5441e+01 mean: 5.9675e-01
[Cossim] L14 cos: 0.997551 max: 7.1728e+01 mean: 7.1910e-01 [Cossim] L14 cos: 0.997547 max: 7.5082e+01 mean: 7.1848e-01
[Cossim] L21 cos: 0.997052 max: 7.5375e+01 mean: 1.3366e+00 [Cossim] L21 cos: 0.997087 max: 7.8800e+01 mean: 1.3215e+00
[Cossim] L27 cos: 0.983419 max: 1.1523e+03 mean: 5.6376e+00 [Cossim] L27 cos: 0.986909 max: 6.4018e+02 mean: 5.5108e+00
[Cossim] Final cos: 0.901694 max: 7.1360e+01 mean: 4.2717e-01 [Cossim] Final cos: 0.904601 max: 7.0254e+01 mean: 4.1660e-01
[Cossim] Logits cos: 0.990653 max: 2.9293e+00 mean: 4.4712e-01 [Cossim] Logits cos: 0.987853 max: 3.3691e+00 mean: 5.0932e-01
[Cossim] NextEmbStep0 cos: 0.999915 max: 3.2489e-03 mean: 7.7490e-04 [Cossim] NextEmbStep0 cos: 0.999915 max: 3.2489e-03 mean: 7.7490e-04
[Cossim] TalkerHiddenStep1 cos: 0.962648 max: 2.3658e+00 mean: 4.7666e-01 [Cossim] TalkerHiddenStep1 cos: 0.963434 max: 2.4030e+00 mean: 4.7170e-01
[Cossim] CodesFull exact: 3.97% (1008 values) [Cossim] CodesFull exact: 2.88% (1008 values)
[Cossim] Audio cos: 0.026175 [Cossim] Audio cos: 0.000079
[Cossim] WAV stft_cos: 0.114082 samples: 120960 [Cossim] WAV stft_cos: 0.080990 samples: 120960
tests/customvoice-CUDA0-Q8_0.log
+19 -19
View File
@@ -40,13 +40,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [41, 2048] first4: 0.021601 -0.009355 0.007161 -0.019744 [Debug] talker-input-embed: [41, 2048] first4: 0.021601 -0.009355 0.007161 -0.019744
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003786 0.008339 -0.005023 -0.000287 [Debug] trailing-text-hidden: [1, 2048] first4: -0.003786 0.008339 -0.005023 -0.000287
[Debug] tts-pad-embed: [2048] first4: -0.003786 0.008339 -0.005023 -0.000287 [Debug] tts-pad-embed: [2048] first4: -0.003786 0.008339 -0.005023 -0.000287
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.385398 -0.059395 0.003227 -0.015955 [Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.385043 -0.059065 0.003278 -0.016107
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -2.095828 0.136483 2.070587 1.095430 [Debug] talker-hidden-prefill-l7: [41, 2048] first4: -2.096979 0.138381 2.068860 1.089124
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.841196 0.297595 2.174365 1.049933 [Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.842276 0.299386 2.172971 1.044518
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -1.422778 0.231152 2.553845 0.556400 [Debug] talker-hidden-prefill-l21: [41, 2048] first4: -1.427377 0.233138 2.551474 0.552877
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -1.260465 -24.832352 -1.697680 25.083447 [Debug] talker-hidden-prefill-l27: [41, 2048] first4: -1.260710 -24.835976 -1.719700 25.117918
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.038955 -0.801399 -0.048752 0.706600 [Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.038966 -0.801593 -0.049389 0.707639
[Debug] talker-logits-prefill: [3072] first4: -0.845418 -7.906702 -2.704062 -4.724893 [Debug] talker-logits-prefill: [3072] first4: -0.833061 -7.949996 -2.577358 -4.753044
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
@@ -65,7 +65,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.039680 -0.032027 -0.001058 0.180065 [Debug] next-emb-step0: [2048] first4: -0.039680 -0.032027 -0.001058 0.180065
[Debug] talker-hidden-step1: [2048] first4: 0.538039 -1.838869 3.074217 -1.235840 [Debug] talker-hidden-step1: [2048] first4: 0.481933 -1.890050 2.987303 -1.150902
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16 [Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
@@ -110,15 +110,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999962 max: 5.1095e-02 mean: 2.0739e-04 [Cossim] Embed cos: 0.999962 max: 5.1095e-02 mean: 2.0739e-04
[Cossim] TrailingText cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05 [Cossim] TrailingText cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05
[Cossim] L0 cos: 0.999986 max: 5.0604e-02 mean: 8.8424e-04 [Cossim] L0 cos: 0.999986 max: 6.3122e-02 mean: 8.8054e-04
[Cossim] L7 cos: 0.999986 max: 1.0962e+01 mean: 4.6588e-02 [Cossim] L7 cos: 0.999985 max: 1.3300e+01 mean: 4.6822e-02
[Cossim] L14 cos: 0.999986 max: 1.0979e+01 mean: 6.0335e-02 [Cossim] L14 cos: 0.999985 max: 1.3311e+01 mean: 6.0435e-02
[Cossim] L21 cos: 0.999981 max: 1.1114e+01 mean: 1.2093e-01 [Cossim] L21 cos: 0.999979 max: 1.3656e+01 mean: 1.2477e-01
[Cossim] L27 cos: 0.999837 max: 8.0113e+01 mean: 5.5753e-01 [Cossim] L27 cos: 0.999803 max: 8.7509e+01 mean: 5.8450e-01
[Cossim] Final cos: 0.998320 max: 1.1710e+01 mean: 4.4568e-02 [Cossim] Final cos: 0.998066 max: 1.3308e+01 mean: 4.6300e-02
[Cossim] Logits cos: 0.999810 max: 3.9394e-01 mean: 7.1349e-02 [Cossim] Logits cos: 0.999815 max: 3.6328e-01 mean: 7.2535e-02
[Cossim] NextEmbStep0 cos: 0.999992 max: 1.1319e-03 mean: 2.3562e-04 [Cossim] NextEmbStep0 cos: 0.999992 max: 1.1319e-03 mean: 2.3562e-04
[Cossim] TalkerHiddenStep1 cos: 0.999540 max: 5.3552e-01 mean: 5.1769e-02 [Cossim] TalkerHiddenStep1 cos: 0.999566 max: 7.0759e-01 mean: 4.9504e-02
[Cossim] CodesFull exact: 5.56% (1008 values) [Cossim] CodesFull exact: 6.65% (1008 values)
[Cossim] Audio cos: 0.089618 [Cossim] Audio cos: 0.078963
[Cossim] WAV stft_cos: 0.180413 samples: 120960 [Cossim] WAV stft_cos: 0.277906 samples: 120960
tests/customvoice-Vulkan0-BF16.log
+96 -1
View File
@@ -16,7 +16,85 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512 [Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend [WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[CausalTransConv] FATAL: 'tok_dec.upsample.0.conv.weight' expected F32 or F16, got type 30 [WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=custom_voice tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=9
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=0, T_ctx=41, hidden=2048, lang=english (id=2050), speaker=vivian (id=3065) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.383264 -0.060895 0.002437 -0.014016
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.342968 0.258215 1.584609 2.188493
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.102334 0.421372 1.683768 2.134429
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.673114 0.328865 2.038729 1.581018
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.925212 -24.483913 -2.209020 26.153484
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.028523 -0.788198 -0.063279 0.734918
[Debug] talker-logits-prefill: [3072] first4: -0.726145 -8.090322 -2.761674 -5.012240
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.039671 -0.032533 -0.000920 0.180072
[Debug] talker-hidden-step1: [2048] first4: 0.521300 -1.852122 3.016847 -1.177232
[Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1191 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=667 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=943 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1177 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1907 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=944 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1711 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=99 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1070 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=194 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000016 0.000020 0.000015 0.000017
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Speaker: vivian [Input] Speaker: vivian
[Input] Lang: english Seed: 42 MaxNewTokens: 64 [Input] Lang: english Seed: 42 MaxNewTokens: 64
@@ -27,3 +105,20 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav [Python] Audio: 120960 samples 24000 Hz 5.04s -> python/customvoice/customvoice-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-customvoice-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf [Quant] BF16 -> ../models/qwen-talker-1.7b-customvoice-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy [GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-customvoice-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --speaker vivian --lang english --max-new 64 --dump cpp/customvoice -o cpp/customvoice/customvoice-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/customvoice/customvoice-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08
[Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] L0 cos: 1.000000 max: 4.7009e-02 mean: 1.1291e-04
[Cossim] L7 cos: 0.999997 max: 2.0094e+01 mean: 1.9638e-03
[Cossim] L14 cos: 0.999997 max: 2.0119e+01 mean: 3.8252e-03
[Cossim] L21 cos: 0.999997 max: 2.0114e+01 mean: 1.1907e-02
[Cossim] L27 cos: 0.999996 max: 1.9367e+01 mean: 6.2017e-02
[Cossim] Final cos: 0.999986 max: 6.4866e-01 mean: 4.6998e-03
[Cossim] Logits cos: 0.999998 max: 4.5917e-02 mean: 7.8233e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 1.3709e-06 mean: 9.5057e-09
[Cossim] TalkerHiddenStep1 cos: 0.999997 max: 3.8364e-02 mean: 4.0905e-03
[Cossim] CodesFull exact: 32.34% (1008 values)
[Cossim] Audio cos: 0.241994
[Cossim] WAV stft_cos: 0.403763 samples: 120960
tests/customvoice-Vulkan0-F32.log
+61 -61
View File
@@ -18,15 +18,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend [WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7 [Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 199.8 MB into backend [WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 199.8 MB [DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 6.0 MB into backend [WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 48.2 MB into backend [WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 48.2 MB [SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend [WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000 [EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 4.0 MB into backend [WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 4.0 MB [EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend [WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB [EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz [Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
@@ -40,48 +40,48 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719 [Debug] talker-input-embed: [41, 2048] first4: 0.021550 -0.009399 0.007126 -0.019719
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409 [Debug] trailing-text-hidden: [1, 2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409 [Debug] tts-pad-embed: [2048] first4: -0.003715 0.008406 -0.005009 -0.000409
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.383174 -0.061029 0.002488 -0.014012 [Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.383174 -0.060999 0.002453 -0.014043
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.341561 0.248239 1.590987 2.199589 [Debug] talker-hidden-prefill-l7: [41, 2048] first4: -1.339357 0.245943 1.590259 2.200182
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.100653 0.411238 1.690403 2.145705 [Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.098146 0.409064 1.689583 2.146009
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.670661 0.321065 2.044533 1.591175 [Debug] talker-hidden-prefill-l21: [41, 2048] first4: -0.668641 0.316357 2.045430 1.590515
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.965589 -24.477230 -2.205596 25.913965 [Debug] talker-hidden-prefill-l27: [41, 2048] first4: -0.973753 -24.432968 -2.196489 25.882349
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.029813 -0.789180 -0.063277 0.729295 [Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.030073 -0.787955 -0.063032 0.728592
[Debug] talker-logits-prefill: [3072] first4: -0.684082 -8.062500 -2.765625 -4.992188 [Debug] talker-logits-prefill: [3072] first4: -0.707031 -8.039062 -2.763672 -4.992188
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1394 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1021 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1572 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1946 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=584 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1686 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1317 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1125 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1139 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=876 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=681 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1118 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=190 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 887.000000 1394.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.066378 -0.021760 -0.024422 -0.054567 [Debug] next-emb-step0: [2048] first4: -0.039671 -0.032533 -0.000920 0.180072
[Debug] talker-hidden-step1: [2048] first4: 0.251683 -6.890096 0.137346 -3.240454 [Debug] talker-hidden-step1: [2048] first4: 0.526778 -1.851193 3.014911 -1.174936
[Sample] step=1 c0=174 u=-1.0000000000 subseq=16 [Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=409 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=149 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=1191 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=648 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=667 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=417 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=943 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=972 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=1177 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=778 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1907 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=274 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=944 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=400 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=641 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1711 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=395 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=99 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1308 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=1070 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=95 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=194 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -91,8 +91,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 887.000000 1394.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000010 0.000010 0.000013 0.000019 [Debug] output-audio: [122880] first4: 0.000015 0.000018 0.000016 0.000016
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -110,15 +110,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08 [Cossim] Embed cos: 1.000000 max: 1.9073e-06 mean: 1.8764e-08
[Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09 [Cossim] TrailingText cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 1.3709e-06 mean: 9.3658e-09
[Cossim] L0 cos: 0.999998 max: 5.8426e-02 mean: 2.1510e-04 [Cossim] L0 cos: 0.999998 max: 5.8426e-02 mean: 2.1676e-04
[Cossim] L7 cos: 0.999996 max: 1.7664e+01 mean: 3.1492e-03 [Cossim] L7 cos: 0.999996 max: 1.7645e+01 mean: 3.1138e-03
[Cossim] L14 cos: 0.999996 max: 1.7679e+01 mean: 5.8671e-03 [Cossim] L14 cos: 0.999996 max: 1.7653e+01 mean: 5.8394e-03
[Cossim] L21 cos: 0.999996 max: 1.7763e+01 mean: 1.7020e-02 [Cossim] L21 cos: 0.999996 max: 1.7698e+01 mean: 1.7566e-02
[Cossim] L27 cos: 0.999995 max: 1.5948e+01 mean: 9.2901e-02 [Cossim] L27 cos: 0.999993 max: 1.8261e+01 mean: 9.6827e-02
[Cossim] Final cos: 0.999976 max: 8.5394e-01 mean: 6.5716e-03 [Cossim] Final cos: 0.999977 max: 7.7382e-01 mean: 6.7446e-03
[Cossim] Logits cos: 0.999991 max: 1.1486e-01 mean: 1.5123e-02 [Cossim] Logits cos: 0.999992 max: 8.7059e-02 mean: 1.3908e-02
[Cossim] NextEmbStep0 cos: 0.456710 max: 5.6155e-01 mean: 5.6235e-02 [Cossim] NextEmbStep0 cos: 1.000000 max: 1.3709e-06 mean: 9.5057e-09
[Cossim] TalkerHiddenStep1 cos: 0.400244 max: 1.1787e+01 mean: 2.0138e+00 [Cossim] TalkerHiddenStep1 cos: 0.999974 max: 1.6781e-01 mean: 1.1962e-02
[Cossim] CodesFull exact: 0.40% (1008 values) [Cossim] CodesFull exact: 17.16% (1008 values)
[Cossim] Audio cos: 0.000580 [Cossim] Audio cos: 0.196257
[Cossim] WAV stft_cos: 0.025550 samples: 120960 [Cossim] WAV stft_cos: 0.313435 samples: 120960
tests/customvoice-Vulkan0-Q4_K_M.log
+54 -54
View File
@@ -40,48 +40,48 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [41, 2048] first4: 0.020406 -0.009921 0.006129 -0.021787 [Debug] talker-input-embed: [41, 2048] first4: 0.020406 -0.009921 0.006129 -0.021787
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003884 0.008386 -0.004502 -0.000786 [Debug] trailing-text-hidden: [1, 2048] first4: -0.003884 0.008386 -0.004502 -0.000786
[Debug] tts-pad-embed: [2048] first4: -0.003884 0.008386 -0.004502 -0.000786 [Debug] tts-pad-embed: [2048] first4: -0.003884 0.008386 -0.004502 -0.000786
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.342387 -0.050952 -0.036901 0.000490 [Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.342387 -0.050922 -0.036901 0.000521
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: 8.684183 -0.420817 6.308812 14.792739 [Debug] talker-hidden-prefill-l7: [41, 2048] first4: 8.707661 -0.417547 6.309919 14.793494
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: 8.668537 -0.144105 6.474887 14.627668 [Debug] talker-hidden-prefill-l14: [41, 2048] first4: 8.692024 -0.140184 6.476104 14.628835
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: 8.614389 0.135494 6.821901 14.258179 [Debug] talker-hidden-prefill-l21: [41, 2048] first4: 8.638645 0.139933 6.822377 14.260940
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: 10.474543 -29.180080 -0.901515 39.380138 [Debug] talker-hidden-prefill-l27: [41, 2048] first4: 10.512966 -29.172842 -0.913447 39.396072
[Debug] talker-hidden-prefill-final: [41, 2048] first4: 0.320889 -0.933492 -0.025663 1.099655 [Debug] talker-hidden-prefill-final: [41, 2048] first4: 0.322031 -0.933157 -0.026000 1.099978
[Debug] talker-logits-prefill: [3072] first4: -0.809082 -8.132812 -1.742188 -5.531250 [Debug] talker-logits-prefill: [3072] first4: -0.792480 -8.101562 -1.723633 -5.464844
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1394 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1050 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=645 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=889 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=417 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=450 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=283 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1306 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=294 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=255 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=791 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=946 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=32 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 887.000000 1394.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.006790 -0.047986 -0.039779 -0.072832 [Debug] next-emb-step0: [2048] first4: -0.041045 -0.031131 0.000245 0.179301
[Debug] talker-hidden-step1: [2048] first4: -0.556227 -6.749060 -0.346863 -2.625830 [Debug] talker-hidden-step1: [2048] first4: 2.269240 -1.724211 2.998043 -1.233295
[Sample] step=1 c0=1028 u=-1.0000000000 subseq=16 [Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1808 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1650 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1843 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1016 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=27 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=656 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1645 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=462 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=672 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=609 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=1518 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=507 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=186 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1721 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=298 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=842 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=1047 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1324 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=733 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=284 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1980 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1632 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=2015 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=1051 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=1760 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=577 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -91,8 +91,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 887.000000 1394.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000006 0.000006 0.000006 0.000005 [Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000011 0.000016
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -110,15 +110,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999698 max: 1.6597e-01 mean: 1.0155e-03 [Cossim] Embed cos: 0.999698 max: 1.6597e-01 mean: 1.0155e-03
[Cossim] TrailingText cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04 [Cossim] TrailingText cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04
[Cossim] TTSPadEmbed cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04 [Cossim] TTSPadEmbed cos: 0.999971 max: 1.2941e-03 mean: 2.2429e-04
[Cossim] L0 cos: 0.999313 max: 4.0368e-01 mean: 6.1748e-03 [Cossim] L0 cos: 0.999313 max: 4.0368e-01 mean: 6.1728e-03
[Cossim] L7 cos: 0.997562 max: 7.2752e+01 mean: 5.9666e-01 [Cossim] L7 cos: 0.997561 max: 7.2753e+01 mean: 5.9674e-01
[Cossim] L14 cos: 0.997544 max: 7.2375e+01 mean: 7.1751e-01 [Cossim] L14 cos: 0.997543 max: 7.2378e+01 mean: 7.1742e-01
[Cossim] L21 cos: 0.997090 max: 7.6005e+01 mean: 1.3161e+00 [Cossim] L21 cos: 0.997095 max: 7.6024e+01 mean: 1.3123e+00
[Cossim] L27 cos: 0.986501 max: 6.5157e+02 mean: 5.4902e+00 [Cossim] L27 cos: 0.986801 max: 6.5177e+02 mean: 5.4749e+00
[Cossim] Final cos: 0.904388 max: 6.9784e+01 mean: 4.1453e-01 [Cossim] Final cos: 0.904293 max: 6.9749e+01 mean: 4.1441e-01
[Cossim] Logits cos: 0.989767 max: 3.2461e+00 mean: 4.6685e-01 [Cossim] Logits cos: 0.989576 max: 3.2573e+00 mean: 4.7012e-01
[Cossim] NextEmbStep0 cos: 0.446799 max: 5.6384e-01 mean: 5.6236e-02 [Cossim] NextEmbStep0 cos: 0.999915 max: 3.2489e-03 mean: 7.7490e-04
[Cossim] TalkerHiddenStep1 cos: 0.424530 max: 1.5663e+01 mean: 1.9080e+00 [Cossim] TalkerHiddenStep1 cos: 0.958074 max: 2.4774e+00 mean: 5.0369e-01
[Cossim] CodesFull exact: 0.30% (1008 values) [Cossim] CodesFull exact: 2.68% (1008 values)
[Cossim] Audio cos: 0.006653 [Cossim] Audio cos: 0.007552
[Cossim] WAV stft_cos: 0.074317 samples: 120960 [Cossim] WAV stft_cos: 0.130484 samples: 120960
tests/customvoice-Vulkan0-Q8_0.log
+54 -54
View File
@@ -40,48 +40,48 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [41, 2048] first4: 0.021601 -0.009355 0.007161 -0.019744 [Debug] talker-input-embed: [41, 2048] first4: 0.021601 -0.009355 0.007161 -0.019744
[Debug] trailing-text-hidden: [1, 2048] first4: -0.003786 0.008339 -0.005023 -0.000287 [Debug] trailing-text-hidden: [1, 2048] first4: -0.003786 0.008339 -0.005023 -0.000287
[Debug] tts-pad-embed: [2048] first4: -0.003786 0.008339 -0.005023 -0.000287 [Debug] tts-pad-embed: [2048] first4: -0.003786 0.008339 -0.005023 -0.000287
[Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.381353 -0.062761 0.001330 -0.016448 [Debug] talker-hidden-prefill-l0: [41, 2048] first4: -0.381353 -0.062791 0.001330 -0.016509
[Debug] talker-hidden-prefill-l7: [41, 2048] first4: -2.105873 0.156029 2.024559 1.129648 [Debug] talker-hidden-prefill-l7: [41, 2048] first4: -2.109648 0.161371 2.017616 1.132363
[Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.855467 0.320307 2.123760 1.083043 [Debug] talker-hidden-prefill-l14: [41, 2048] first4: -1.859596 0.326090 2.117046 1.085369
[Debug] talker-hidden-prefill-l21: [41, 2048] first4: -1.445384 0.256552 2.501003 0.579517 [Debug] talker-hidden-prefill-l21: [41, 2048] first4: -1.449659 0.263131 2.492201 0.581436
[Debug] talker-hidden-prefill-l27: [41, 2048] first4: -1.413932 -24.682590 -1.774915 24.908796 [Debug] talker-hidden-prefill-l27: [41, 2048] first4: -1.422006 -24.676947 -1.766300 24.876963
[Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.043676 -0.796183 -0.050946 0.701342 [Debug] talker-hidden-prefill-final: [41, 2048] first4: -0.043907 -0.795666 -0.050677 0.700152
[Debug] talker-logits-prefill: [3072] first4: -0.807129 -7.988281 -2.699219 -4.851562 [Debug] talker-logits-prefill: [3072] first4: -0.840332 -8.007812 -2.685547 -4.832031
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=355 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1394 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=22 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1021 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=1174 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1572 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=1093 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1946 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=625 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=584 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1814 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1686 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1058 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1317 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=905 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1125 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=1846 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=1139 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=1247 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=876 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=681 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=889 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1118 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=190 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=901 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1159.000000 887.000000 1394.000000 [Debug] codes-step0: [16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] next-emb-step0: [2048] first4: -0.066422 -0.021942 -0.023905 -0.054010 [Debug] next-emb-step0: [2048] first4: -0.039680 -0.032027 -0.001058 0.180065
[Debug] talker-hidden-step1: [2048] first4: 0.161350 -6.832321 0.084924 -3.220334 [Debug] talker-hidden-step1: [2048] first4: 0.569267 -1.901817 3.022171 -1.198206
[Sample] step=1 c0=174 u=-1.0000000000 subseq=16 [Sample] step=1 c0=259 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1159 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=259 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=887 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=1160 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=409 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1605 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=149 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=1191 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=648 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=667 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1614 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=417 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=943 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=972 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=1177 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=778 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1907 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=274 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=944 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=400 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=126 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=641 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=1711 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=395 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=99 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1308 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=1070 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=171 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=194 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -91,8 +91,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 887.000000 1394.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1159.000000 355.000000 22.000000
[Debug] output-audio: [122880] first4: 0.000011 0.000011 0.000013 0.000018 [Debug] output-audio: [122880] first4: 0.000015 0.000017 0.000015 0.000017
[WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/customvoice/customvoice-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/customvoice/customvoice-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -110,15 +110,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999962 max: 5.1095e-02 mean: 2.0739e-04 [Cossim] Embed cos: 0.999962 max: 5.1095e-02 mean: 2.0739e-04
[Cossim] TrailingText cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05 [Cossim] TrailingText cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 1.8678e-04 mean: 4.3506e-05
[Cossim] L0 cos: 0.999990 max: 5.0900e-02 mean: 6.5162e-04 [Cossim] L0 cos: 0.999990 max: 5.8643e-02 mean: 6.5001e-04
[Cossim] L7 cos: 0.999982 max: 2.1600e+01 mean: 4.4796e-02 [Cossim] L7 cos: 0.999982 max: 2.5612e+01 mean: 4.4817e-02
[Cossim] L14 cos: 0.999982 max: 2.1531e+01 mean: 5.5236e-02 [Cossim] L14 cos: 0.999982 max: 2.5547e+01 mean: 5.5255e-02
[Cossim] L21 cos: 0.999979 max: 2.1861e+01 mean: 1.0253e-01 [Cossim] L21 cos: 0.999979 max: 2.5867e+01 mean: 1.0312e-01
[Cossim] L27 cos: 0.999892 max: 6.5629e+01 mean: 4.4925e-01 [Cossim] L27 cos: 0.999883 max: 6.8940e+01 mean: 4.5189e-01
[Cossim] Final cos: 0.998972 max: 1.0132e+01 mean: 3.4602e-02 [Cossim] Final cos: 0.998955 max: 1.0042e+01 mean: 3.4832e-02
[Cossim] Logits cos: 0.999883 max: 3.0927e-01 mean: 5.9168e-02 [Cossim] Logits cos: 0.999876 max: 3.4833e-01 mean: 6.0124e-02
[Cossim] NextEmbStep0 cos: 0.456857 max: 5.6147e-01 mean: 5.6227e-02 [Cossim] NextEmbStep0 cos: 0.999992 max: 1.1319e-03 mean: 2.3562e-04
[Cossim] TalkerHiddenStep1 cos: 0.397042 max: 1.1716e+01 mean: 2.0233e+00 [Cossim] TalkerHiddenStep1 cos: 0.999773 max: 3.4846e-01 mean: 3.6412e-02
[Cossim] CodesFull exact: 0.30% (1008 values) [Cossim] CodesFull exact: 7.84% (1008 values)
[Cossim] Audio cos: 0.002140 [Cossim] Audio cos: 0.063656
[Cossim] WAV stft_cos: 0.038675 samples: 120960 [Cossim] WAV stft_cos: 0.318519 samples: 120960
tests/tts-CPU-BF16.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-voicedesign-BF16.gguf: 404 tensors, data at offset 5949344
[WeightCtx] Loaded 316 tensors, 3322.0 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 334.1 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-BF16.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=voice_design tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=12, T_ctx=52, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [52, 2048] first4: 0.018740 -0.007903 0.004584 -0.013295
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.377323 -0.053064 -0.039823 -0.054050
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.917138 0.543871 0.929805 0.987515
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.696294 0.739704 1.047741 0.896942
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.292889 0.669339 1.303152 0.481775
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.671435 -28.812714 -2.738419 31.965797
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.302407 -1.363576 -0.114833 1.321304
[Debug] talker-logits-prefill: [3072] first4: -3.017047 -10.117426 -1.564162 -3.028543
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1703 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1047 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=37 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1759 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1163 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=564 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=796 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=783 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1999 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.004818 -0.029755 0.017770 0.080159
[Debug] talker-hidden-step1: [2048] first4: 0.561001 -0.633637 7.049995 -3.812517
[Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1879 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=628 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1799 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=908 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=132 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=942 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1522 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=317 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=623 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] output-audio: [122880] first4: 0.000048 0.000059 0.000047 0.000050
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Instruct: male, young adult, moderate pitch
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] InstructIds shape: (1, 12)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-voicedesign-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/tts/tts-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08
[Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] L0 cos: 1.000000 max: 3.8467e-02 mean: 1.3029e-04
[Cossim] L7 cos: 0.999999 max: 9.2734e+00 mean: 1.5558e-03
[Cossim] L14 cos: 0.999999 max: 9.2463e+00 mean: 2.9746e-03
[Cossim] L21 cos: 0.999999 max: 9.2458e+00 mean: 8.5830e-03
[Cossim] L27 cos: 0.999996 max: 2.2716e+01 mean: 4.4241e-02
[Cossim] Final cos: 0.999993 max: 4.5066e-01 mean: 4.8795e-03
[Cossim] Logits cos: 0.999995 max: 5.1050e-02 mean: 8.5875e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 8.3447e-07 mean: 9.2208e-09
[Cossim] TalkerHiddenStep1 cos: 0.999991 max: 4.3530e-02 mean: 9.3439e-03
[Cossim] CodesFull exact: 12.20% (1008 values)
[Cossim] Audio cos: 0.208068
[Cossim] WAV stft_cos: 0.533081 samples: 120960
tests/tts-CPU-F32.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-voicedesign-F32.gguf: 404 tensors, data at offset 5949248
[WeightCtx] Loaded 316 tensors, 6643.5 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 668.1 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-F32.gguf: 398 tensors, data at offset 29664
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 116.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=voice_design tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=12, T_ctx=52, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [52, 2048] first4: 0.018740 -0.007903 0.004584 -0.013295
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.377575 -0.053131 -0.039824 -0.054256
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.919110 0.546713 0.930029 0.992751
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.698142 0.742027 1.047813 0.901880
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.293931 0.674182 1.303767 0.486382
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.640283 -28.858608 -2.782684 32.034195
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.300681 -1.364322 -0.116567 1.322749
[Debug] talker-logits-prefill: [3072] first4: -3.029264 -10.119155 -1.570520 -3.061079
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1703 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1047 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=37 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1759 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1163 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=564 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=796 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=783 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1999 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.004818 -0.029755 0.017770 0.080159
[Debug] talker-hidden-step1: [2048] first4: 0.550206 -0.654530 7.058207 -3.804737
[Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1879 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=628 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1799 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=908 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=132 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=942 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1522 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=317 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=623 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] output-audio: [122880] first4: 0.000047 0.000055 0.000044 0.000047
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Instruct: male, young adult, moderate pitch
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] InstructIds shape: (1, 12)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav
[Quant] F32 -> ../models/qwen-talker-1.7b-voicedesign-F32.gguf + ../models/qwen-tokenizer-12hz-F32.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-F32.gguf --codec ../models/qwen-tokenizer-12hz-F32.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/tts/tts-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08
[Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] L0 cos: 1.000000 max: 7.6294e-06 mean: 7.3912e-08
[Cossim] L7 cos: 1.000000 max: 2.6855e-03 mean: 4.4471e-07
[Cossim] L14 cos: 1.000000 max: 2.6855e-03 mean: 9.0014e-07
[Cossim] L21 cos: 1.000000 max: 2.6855e-03 mean: 2.5103e-06
[Cossim] L27 cos: 1.000000 max: 4.1504e-03 mean: 1.1344e-05
[Cossim] Final cos: 1.000000 max: 1.0681e-04 mean: 1.3017e-06
[Cossim] Logits cos: 1.000000 max: 2.3127e-05 mean: 2.3441e-06
[Cossim] NextEmbStep0 cos: 1.000000 max: 8.3447e-07 mean: 9.2208e-09
[Cossim] TalkerHiddenStep1 cos: 1.000000 max: 2.6703e-05 mean: 2.6897e-06
[Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 1.000000
[Cossim] WAV stft_cos: 1.000000 samples: 120960
tests/tts-CPU-Q4_K_M.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-voicedesign-Q4_K_M.gguf: 404 tensors, data at offset 5949344
[WeightCtx] Loaded 316 tensors, 1012.7 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 109.5 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-Q4_K_M.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 16.4 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 25.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 13.6 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=voice_design tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=12, T_ctx=52, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [52, 2048] first4: 0.017779 -0.008618 0.002377 -0.014714
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004540 0.007622 -0.004462 -0.000982
[Debug] tts-pad-embed: [2048] first4: -0.004540 0.007622 -0.004462 -0.000982
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.362450 -0.021774 -0.019305 -0.088981
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -11.138859 0.700413 5.853215 -2.349519
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -10.711702 0.626058 5.997408 -2.466321
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -9.763048 -0.108736 6.836337 -1.704187
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -13.079835 -29.332205 7.916404 29.596735
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.614709 -1.439244 0.344182 1.268398
[Debug] talker-logits-prefill: [3072] first4: -6.182429 -8.585429 0.276684 -0.180746
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=2030 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=6 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=2026 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1751 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1643 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1822 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=86 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=288 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=248 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1713 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1860 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] next-emb-step0: [2048] first4: -0.010564 -0.042134 0.031385 0.081208
[Debug] talker-hidden-step1: [2048] first4: 1.453258 -2.064452 3.893206 -3.106290
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1782 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=515 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1210 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=278 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1579 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1902 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1611 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1625 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=324 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=903 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=706 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1759 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] output-audio: [122880] first4: 0.000076 0.000101 0.000098 0.000093
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Instruct: male, young adult, moderate pitch
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] InstructIds shape: (1, 12)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav
[Quant] Q4_K_M -> ../models/qwen-talker-1.7b-voicedesign-Q4_K_M.gguf + ../models/qwen-tokenizer-12hz-Q4_K_M.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-Q4_K_M.gguf --codec ../models/qwen-tokenizer-12hz-Q4_K_M.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/tts/tts-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 0.999680 max: 1.3213e-02 mean: 9.5199e-04
[Cossim] TrailingText cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] L0 cos: 0.999123 max: 1.1177e+00 mean: 6.7299e-03
[Cossim] L7 cos: 0.997435 max: 1.6900e+02 mean: 4.3776e-01
[Cossim] L14 cos: 0.997407 max: 1.6971e+02 mean: 5.4610e-01
[Cossim] L21 cos: 0.996991 max: 1.6755e+02 mean: 1.0205e+00
[Cossim] L27 cos: 0.985533 max: 6.2395e+02 mean: 4.7228e+00
[Cossim] Final cos: 0.934666 max: 2.7791e+01 mean: 5.1024e-01
[Cossim] Logits cos: 0.878809 max: 9.5592e+00 mean: 1.8377e+00
[Cossim] NextEmbStep0 cos: 0.798012 max: 1.6584e-01 mean: 3.4465e-02
[Cossim] TalkerHiddenStep1 cos: 0.676446 max: 1.0832e+01 mean: 1.8567e+00
[Cossim] CodesFull exact: 0.99% (1008 values)
[Cossim] Audio cos: 0.011661
[Cossim] WAV stft_cos: 0.199112 samples: 120960
tests/tts-CPU-Q8_0.log
+125
View File
@@ -0,0 +1,125 @@
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 97247 MiB):
Device 0: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, compute capability 12.0, VMM: yes, VRAM: 97247 MiB
load_backend: loaded CUDA backend from /mnt/workspace/qwentts.cpp/build/libggml-cuda.so
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA RTX PRO 6000 Blackwell Workstation Edition (NVIDIA) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from /mnt/workspace/qwentts.cpp/build/libggml-vulkan.so
load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-cpu-zen4.so
[Load] Talker backend: CPU (CPU threads: 16)
[GGUF] ../models/qwen-talker-1.7b-voicedesign-Q8_0.gguf: 404 tensors, data at offset 5949344
[WeightCtx] Loaded 316 tensors, 1765.0 MB into backend
[Talker] Loaded: 28 layers, hidden 2048, heads 16/8, head_dim 128, FFN 6144, RoPE theta 1000000, mrope sections [24,20,20] interleaved=0
[WeightCtx] Loaded 88 tensors, 177.5 MB into backend
[CodePredictor] Loaded: 5 layers, hidden 1024, heads 16/8, head_dim 128, FFN 3072, RoPE theta 1000000, 15 acoustic codebooks (vocab 2048 each), mtp_proj linear
[GGUF] ../models/qwen-tokenizer-12hz-Q8_0.gguf: 398 tensors, data at offset 29760
[WeightCtx] Loaded 18 tensors, 33.0 MB into backend
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 30.9 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 33.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 25.6 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=voice_design tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=12, T_ctx=52, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [52, 2048] first4: 0.018647 -0.007834 0.004680 -0.013328
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004554 0.007647 -0.004656 -0.000707
[Debug] tts-pad-embed: [2048] first4: -0.004554 0.007647 -0.004656 -0.000707
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.383844 -0.050547 -0.040529 -0.050115
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -1.392936 0.381311 1.521645 0.554877
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -1.165570 0.579111 1.650208 0.467172
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.749908 0.475319 1.954961 0.103869
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.858568 -29.632925 -2.005177 31.725758
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.308463 -1.391446 -0.083429 1.301145
[Debug] talker-logits-prefill: [3072] first4: -2.998427 -9.951941 -1.629458 -2.914670
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=2030 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1378 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1228 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1751 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=308 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=1479 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1673 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=625 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=146 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=1108 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1546 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1569 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=381 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] next-emb-step0: [2048] first4: 0.039071 -0.102078 -0.021939 0.058714
[Debug] talker-hidden-step1: [2048] first4: 1.026723 -0.054218 7.733221 -6.603862
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1782 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1871 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1042 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=804 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=493 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1579 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1886 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1857 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=86 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=1657 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1967 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=706 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=344 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=82 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=1435 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] output-audio: [122880] first4: 0.000076 0.000096 0.000082 0.000083
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Instruct: male, young adult, moderate pitch
[Input] Lang: english Seed: 42 MaxNewTokens: 64
[Input] Mode: greedy
[Python] Device: cuda
[Python] InputIds shape: (1, 38)
[Python] InstructIds shape: (1, 12)
[Python] Codes shape: (63, 16) (T_frames, num_code_groups)
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav
[Quant] Q8_0 -> ../models/qwen-talker-1.7b-voicedesign-Q8_0.gguf + ../models/qwen-tokenizer-12hz-Q8_0.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-Q8_0.gguf --codec ../models/qwen-tokenizer-12hz-Q8_0.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/tts/tts-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 0.999989 max: 2.2227e-03 mean: 1.6673e-04
[Cossim] TrailingText cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05
[Cossim] L0 cos: 0.999985 max: 3.6777e-02 mean: 9.2539e-04
[Cossim] L7 cos: 0.999987 max: 8.6318e+00 mean: 3.3859e-02
[Cossim] L14 cos: 0.999987 max: 8.7617e+00 mean: 4.4713e-02
[Cossim] L21 cos: 0.999983 max: 8.5483e+00 mean: 9.0651e-02
[Cossim] L27 cos: 0.999900 max: 1.2904e+02 mean: 4.0550e-01
[Cossim] Final cos: 0.999441 max: 2.6548e+00 mean: 4.7593e-02
[Cossim] Logits cos: 0.999824 max: 3.2616e-01 mean: 5.4241e-02
[Cossim] NextEmbStep0 cos: 0.791728 max: 1.7376e-01 mean: 3.3813e-02
[Cossim] TalkerHiddenStep1 cos: 0.820984 max: 9.8897e+00 mean: 1.5116e+00
[Cossim] CodesFull exact: 0.79% (1008 values)
[Cossim] Audio cos: 0.039976
[Cossim] WAV stft_cos: 0.224882 samples: 120960
tests/tts-CUDA0-BF16.log
+96 -1
View File
@@ -16,7 +16,85 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512 [Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend [WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[CausalTransConv] FATAL: 'tok_dec.upsample.0.conv.weight' expected F32 or F16, got type 30 [WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=voice_design tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=12, T_ctx=52, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [52, 2048] first4: 0.018740 -0.007903 0.004584 -0.013295
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.378232 -0.052702 -0.040521 -0.054066
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.920871 0.547279 0.941228 0.993006
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.699448 0.743224 1.058581 0.902093
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.292916 0.676742 1.315608 0.487073
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.728775 -28.810892 -2.702032 32.115330
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.304049 -1.359213 -0.112952 1.323321
[Debug] talker-logits-prefill: [3072] first4: -3.093750 -10.125000 -1.593750 -3.062500
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1703 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1047 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=37 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1759 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1163 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=564 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=796 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=783 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1999 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.004818 -0.029755 0.017770 0.080159
[Debug] talker-hidden-step1: [2048] first4: 0.532398 -0.664292 7.064521 -3.802885
[Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1879 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=628 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1799 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=908 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=132 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=942 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1522 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=317 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=623 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] output-audio: [122880] first4: 0.000048 0.000058 0.000047 0.000050
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Instruct: male, young adult, moderate pitch [Input] Instruct: male, young adult, moderate pitch
[Input] Lang: english Seed: 42 MaxNewTokens: 64 [Input] Lang: english Seed: 42 MaxNewTokens: 64
@@ -28,3 +106,20 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav [Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-voicedesign-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf [Quant] BF16 -> ../models/qwen-talker-1.7b-voicedesign-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy [GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/tts/tts-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08
[Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] L0 cos: 0.999996 max: 1.6262e-01 mean: 2.9447e-04
[Cossim] L7 cos: 0.999996 max: 1.4347e+01 mean: 2.8592e-03
[Cossim] L14 cos: 0.999996 max: 1.4347e+01 mean: 5.5062e-03
[Cossim] L21 cos: 0.999996 max: 1.4373e+01 mean: 1.6051e-02
[Cossim] L27 cos: 0.999990 max: 3.0361e+01 mean: 8.5810e-02
[Cossim] Final cos: 0.999974 max: 6.1434e-01 mean: 9.3876e-03
[Cossim] Logits cos: 0.999979 max: 1.0430e-01 mean: 2.0474e-02
[Cossim] NextEmbStep0 cos: 1.000000 max: 8.3447e-07 mean: 9.2208e-09
[Cossim] TalkerHiddenStep1 cos: 0.999992 max: 6.7879e-02 mean: 8.6519e-03
[Cossim] CodesFull exact: 23.02% (1008 values)
[Cossim] Audio cos: 0.408996
[Cossim] WAV stft_cos: 0.714920 samples: 120960
tests/tts-CUDA0-F32.log
+23 -23
View File
@@ -18,15 +18,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend [WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7 [Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 199.8 MB into backend [WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 199.8 MB [DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 6.0 MB into backend [WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 48.2 MB into backend [WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 48.2 MB [SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend [WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000 [EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 4.0 MB into backend [WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 4.0 MB [EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend [WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB [EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz [Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
@@ -40,13 +40,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [52, 2048] first4: 0.018740 -0.007903 0.004584 -0.013295 [Debug] talker-input-embed: [52, 2048] first4: 0.018740 -0.007903 0.004584 -0.013295
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724 [Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724 [Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.377575 -0.053131 -0.039824 -0.054256 [Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.377584 -0.053122 -0.039813 -0.054261
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.919110 0.546713 0.930030 0.992750 [Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.919122 0.546647 0.930051 0.992817
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.698142 0.742027 1.047814 0.901879 [Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.698185 0.741910 1.047810 0.901926
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.293931 0.674182 1.303768 0.486382 [Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.293857 0.673920 1.303763 0.486586
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.640275 -28.858603 -2.782673 32.034184 [Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.640599 -28.856728 -2.781946 32.033070
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.300680 -1.364321 -0.116567 1.322748 [Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.300702 -1.364263 -0.116539 1.322731
[Debug] talker-logits-prefill: [3072] first4: -3.029268 -10.119147 -1.570518 -3.061075 [Debug] talker-logits-prefill: [3072] first4: -3.029830 -10.119137 -1.570917 -3.058243
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
@@ -65,7 +65,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000 [Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.004818 -0.029755 0.017770 0.080159 [Debug] next-emb-step0: [2048] first4: 0.004818 -0.029755 0.017770 0.080159
[Debug] talker-hidden-step1: [2048] first4: 0.550205 -0.654532 7.058204 -3.804739 [Debug] talker-hidden-step1: [2048] first4: 0.551770 -0.653224 7.056627 -3.804871
[Sample] step=1 c0=2042 u=-1.0000000000 subseq=16 [Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08 [Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08
[Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09 [Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] L0 cos: 1.000000 max: 2.2888e-05 mean: 6.9453e-08 [Cossim] L0 cos: 1.000000 max: 1.8311e-03 mean: 2.0342e-05
[Cossim] L7 cos: 1.000000 max: 4.8828e-04 mean: 3.9031e-07 [Cossim] L7 cos: 1.000000 max: 5.3491e-01 mean: 1.6043e-04
[Cossim] L14 cos: 1.000000 max: 4.8828e-04 mean: 9.1018e-07 [Cossim] L14 cos: 1.000000 max: 5.3369e-01 mean: 3.9851e-04
[Cossim] L21 cos: 1.000000 max: 4.8828e-04 mean: 2.7898e-06 [Cossim] L21 cos: 1.000000 max: 5.2856e-01 mean: 1.1915e-03
[Cossim] L27 cos: 1.000000 max: 3.4180e-03 mean: 1.0663e-05 [Cossim] L27 cos: 1.000000 max: 9.4678e-01 mean: 3.8733e-03
[Cossim] Final cos: 1.000000 max: 2.3651e-04 mean: 1.3797e-06 [Cossim] Final cos: 1.000000 max: 6.4301e-02 mean: 5.7386e-04
[Cossim] Logits cos: 1.000000 max: 1.7285e-05 mean: 2.6001e-06 [Cossim] Logits cos: 1.000000 max: 7.1373e-03 mean: 1.0775e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 8.3447e-07 mean: 9.2208e-09 [Cossim] NextEmbStep0 cos: 1.000000 max: 8.3447e-07 mean: 9.2208e-09
[Cossim] TalkerHiddenStep1 cos: 1.000000 max: 1.7166e-05 mean: 2.3196e-06 [Cossim] TalkerHiddenStep1 cos: 1.000000 max: 7.6799e-03 mean: 1.0526e-03
[Cossim] CodesFull exact: 100.00% (1008 values) [Cossim] CodesFull exact: 100.00% (1008 values)
[Cossim] Audio cos: 0.999998 [Cossim] Audio cos: 0.999998
[Cossim] WAV stft_cos: 0.999999 samples: 120960 [Cossim] WAV stft_cos: 0.999999 samples: 120960
tests/tts-CUDA0-Q4_K_M.log
+43 -43
View File
@@ -40,46 +40,46 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [52, 2048] first4: 0.017779 -0.008618 0.002377 -0.014714 [Debug] talker-input-embed: [52, 2048] first4: 0.017779 -0.008618 0.002377 -0.014714
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004540 0.007622 -0.004462 -0.000982 [Debug] trailing-text-hidden: [1, 2048] first4: -0.004540 0.007622 -0.004462 -0.000982
[Debug] tts-pad-embed: [2048] first4: -0.004540 0.007622 -0.004462 -0.000982 [Debug] tts-pad-embed: [2048] first4: -0.004540 0.007622 -0.004462 -0.000982
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.371689 -0.023940 -0.019825 -0.090360 [Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.372901 -0.024501 -0.020248 -0.090314
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -11.135981 0.675024 5.790349 -2.381296 [Debug] talker-hidden-prefill-l7: [52, 2048] first4: -11.007947 0.717153 5.867706 -2.264582
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -10.706117 0.605723 5.933682 -2.483038 [Debug] talker-hidden-prefill-l14: [52, 2048] first4: -10.569766 0.656971 6.022027 -2.353054
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -9.700500 -0.234456 6.598506 -1.840014 [Debug] talker-hidden-prefill-l21: [52, 2048] first4: -9.637493 -0.210011 6.651770 -1.720735
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -13.985462 -29.550419 6.957130 29.017033 [Debug] talker-hidden-prefill-l27: [52, 2048] first4: -13.348853 -29.921713 7.442925 29.507101
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.661750 -1.459834 0.304538 1.252031 [Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.630546 -1.475644 0.325244 1.270995
[Debug] talker-logits-prefill: [3072] first4: -6.002866 -8.173971 0.543856 0.185978 [Debug] talker-logits-prefill: [3072] first4: -6.003940 -8.301048 0.621289 0.166484
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=2030 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=2030 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=6 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=6 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=2026 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=2026 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1751 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=359 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=99 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1643 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=971 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1822 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1221 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1625 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=288 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=248 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=437 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=706 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1384 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1759 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=1484 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=812 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1008 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=1435 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=1409 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 2030.000000 [Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] next-emb-step0: [2048] first4: 0.052074 -0.063247 0.048163 0.061720 [Debug] next-emb-step0: [2048] first4: 0.034390 -0.060524 0.012125 0.060673
[Debug] talker-hidden-step1: [2048] first4: 0.974696 -1.231946 4.508642 -3.423083 [Debug] talker-hidden-step1: [2048] first4: 1.072873 -2.218389 4.078456 -2.675533
[Sample] step=1 c0=215 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1782 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=85 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1336 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=282 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1531 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=278 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=286 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1155 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1579 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1902 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=370 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1440 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=1440 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1124 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=86 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=324 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=243 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1331 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=903 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=914 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=996 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1759 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=344 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1772 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=1772 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
@@ -92,7 +92,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 2030.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] output-audio: [122880] first4: 0.000106 0.000147 0.000150 0.000143 [Debug] output-audio: [122880] first4: 0.000066 0.000088 0.000087 0.000084
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999680 max: 1.3213e-02 mean: 9.5199e-04 [Cossim] Embed cos: 0.999680 max: 1.3213e-02 mean: 9.5199e-04
[Cossim] TrailingText cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04 [Cossim] TrailingText cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04 [Cossim] TTSPadEmbed cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] L0 cos: 0.999158 max: 8.3527e-01 mean: 6.7138e-03 [Cossim] L0 cos: 0.999153 max: 8.2018e-01 mean: 6.7328e-03
[Cossim] L7 cos: 0.997488 max: 1.3347e+02 mean: 4.3744e-01 [Cossim] L7 cos: 0.997483 max: 1.4311e+02 mean: 4.3763e-01
[Cossim] L14 cos: 0.997461 max: 1.3464e+02 mean: 5.4416e-01 [Cossim] L14 cos: 0.997455 max: 1.4454e+02 mean: 5.4431e-01
[Cossim] L21 cos: 0.997052 max: 1.3294e+02 mean: 1.0180e+00 [Cossim] L21 cos: 0.997026 max: 1.4280e+02 mean: 1.0254e+00
[Cossim] L27 cos: 0.988593 max: 5.2666e+02 mean: 4.5783e+00 [Cossim] L27 cos: 0.988444 max: 5.2529e+02 mean: 4.5876e+00
[Cossim] Final cos: 0.929912 max: 2.7847e+01 mean: 5.2305e-01 [Cossim] Final cos: 0.932398 max: 3.1110e+01 mean: 5.1677e-01
[Cossim] Logits cos: 0.883872 max: 9.1003e+00 mean: 1.7283e+00 [Cossim] Logits cos: 0.879634 max: 9.2089e+00 mean: 1.7576e+00
[Cossim] NextEmbStep0 cos: 0.784060 max: 1.5974e-01 mean: 3.5553e-02 [Cossim] NextEmbStep0 cos: 0.824996 max: 1.5647e-01 mean: 3.1773e-02
[Cossim] TalkerHiddenStep1 cos: 0.693917 max: 9.2592e+00 mean: 1.8013e+00 [Cossim] TalkerHiddenStep1 cos: 0.688264 max: 1.1933e+01 mean: 1.8067e+00
[Cossim] CodesFull exact: 0.30% (1008 values) [Cossim] CodesFull exact: 0.60% (1008 values)
[Cossim] Audio cos: 0.013818 [Cossim] Audio cos: 0.007003
[Cossim] WAV stft_cos: 0.143592 samples: 120960 [Cossim] WAV stft_cos: 0.240854 samples: 120960
tests/tts-CUDA0-Q8_0.log
+25 -25
View File
@@ -40,13 +40,13 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] talker-input-embed: [52, 2048] first4: 0.018647 -0.007834 0.004680 -0.013328 [Debug] talker-input-embed: [52, 2048] first4: 0.018647 -0.007834 0.004680 -0.013328
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004554 0.007647 -0.004656 -0.000707 [Debug] trailing-text-hidden: [1, 2048] first4: -0.004554 0.007647 -0.004656 -0.000707
[Debug] tts-pad-embed: [2048] first4: -0.004554 0.007647 -0.004656 -0.000707 [Debug] tts-pad-embed: [2048] first4: -0.004554 0.007647 -0.004656 -0.000707
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.383302 -0.050605 -0.040526 -0.049895 [Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.383292 -0.050999 -0.040456 -0.049748
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -1.393858 0.381871 1.525452 0.556245 [Debug] talker-hidden-prefill-l7: [52, 2048] first4: -1.392987 0.381949 1.519604 0.554772
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -1.167568 0.580672 1.653311 0.469169 [Debug] talker-hidden-prefill-l14: [52, 2048] first4: -1.165597 0.579776 1.647923 0.467421
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.757229 0.472590 1.960911 0.103175 [Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.756077 0.475114 1.955364 0.100142
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.869149 -29.638502 -2.016915 31.724474 [Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.881919 -29.637997 -2.021209 31.717108
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.308995 -1.391961 -0.083932 1.301329 [Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.309583 -1.392001 -0.084115 1.301087
[Debug] talker-logits-prefill: [3072] first4: -2.926836 -10.031683 -1.534877 -3.103800 [Debug] talker-logits-prefill: [3072] first4: -2.904243 -10.136507 -1.713597 -3.231483
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
@@ -65,7 +65,7 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000 [Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.004793 -0.029425 0.017809 0.080303 [Debug] next-emb-step0: [2048] first4: 0.004793 -0.029425 0.017809 0.080303
[Debug] talker-hidden-step1: [2048] first4: 0.518513 -0.706861 6.929946 -3.851983 [Debug] talker-hidden-step1: [2048] first4: 0.576312 -0.721774 7.087047 -3.985323
[Sample] step=1 c0=2042 u=-1.0000000000 subseq=16 [Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
@@ -76,12 +76,12 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=943 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=908 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1520 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=132 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=958 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=942 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=772 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1522 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1728 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=317 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=1 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=1202 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999989 max: 2.2227e-03 mean: 1.6673e-04 [Cossim] Embed cos: 0.999989 max: 2.2227e-03 mean: 1.6673e-04
[Cossim] TrailingText cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05 [Cossim] TrailingText cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05
[Cossim] L0 cos: 0.999985 max: 4.3993e-02 mean: 9.2738e-04 [Cossim] L0 cos: 0.999986 max: 3.5740e-02 mean: 9.2337e-04
[Cossim] L7 cos: 0.999987 max: 9.9563e+00 mean: 3.3943e-02 [Cossim] L7 cos: 0.999987 max: 1.0577e+01 mean: 3.4083e-02
[Cossim] L14 cos: 0.999987 max: 9.9055e+00 mean: 4.4787e-02 [Cossim] L14 cos: 0.999987 max: 1.0719e+01 mean: 4.4989e-02
[Cossim] L21 cos: 0.999983 max: 9.6638e+00 mean: 8.9751e-02 [Cossim] L21 cos: 0.999983 max: 1.0483e+01 mean: 9.0445e-02
[Cossim] L27 cos: 0.999887 max: 1.3903e+02 mean: 4.0896e-01 [Cossim] L27 cos: 0.999907 max: 1.3287e+02 mean: 3.9520e-01
[Cossim] Final cos: 0.999444 max: 3.9794e+00 mean: 4.7093e-02 [Cossim] Final cos: 0.999468 max: 2.6326e+00 mean: 4.6949e-02
[Cossim] Logits cos: 0.999819 max: 3.5383e-01 mean: 5.5424e-02 [Cossim] Logits cos: 0.999799 max: 3.3817e-01 mean: 6.3672e-02
[Cossim] NextEmbStep0 cos: 0.999990 max: 1.0252e-03 mean: 2.4056e-04 [Cossim] NextEmbStep0 cos: 0.999990 max: 1.0252e-03 mean: 2.4056e-04
[Cossim] TalkerHiddenStep1 cos: 0.999389 max: 4.4522e-01 mean: 7.6610e-02 [Cossim] TalkerHiddenStep1 cos: 0.999445 max: 5.1719e-01 mean: 7.3323e-02
[Cossim] CodesFull exact: 4.17% (1008 values) [Cossim] CodesFull exact: 4.46% (1008 values)
[Cossim] Audio cos: 0.000338 [Cossim] Audio cos: 0.017067
[Cossim] WAV stft_cos: 0.300761 samples: 120960 [Cossim] WAV stft_cos: 0.253108 samples: 120960
tests/tts-Vulkan0-BF16.log
+96 -1
View File
@@ -16,7 +16,85 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512 [Quantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512
[WeightCtx] Loaded 93 tensors, 58.1 MB into backend [WeightCtx] Loaded 93 tensors, 58.1 MB into backend
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[CausalTransConv] FATAL: 'tok_dec.upsample.0.conv.weight' expected F32 or F16, got type 30 [WeightCtx] Loaded 22 tensors, 48.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 48.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
[KVCache] Allocated: 28 layers, 8 KV heads, head_dim 128, max_seq_len 4096 -> 896 MB
[KVCache] Allocated: 5 layers, 8 KV heads, head_dim 128, max_seq_len 16 -> 0 MB
[Pipeline] Loaded: arch=1b7 variant=voice_design tokenizer=qwen3_tts_tokenizer_12hz codebooks=16 speaker_encoder=absent speakers=0
[BPE] Loaded from GGUF: 151676 vocab, 151291 merges, eos_id=151643
[BPE] Registered 5 arch special tokens (total specials=6)
[Prompt] Built: 38 ids, N_text=30, N_instruct=12, T_ctx=52, hidden=2048, lang=english (id=2050), speaker=none (id=-1) ref_spk_emb=no icl=no
[Debug] prompt-ids: [38] first4: 151644.000000 77091.000000 198.000000 80.000000
[Debug] talker-input-embed: [52, 2048] first4: 0.018740 -0.007903 0.004584 -0.013295
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.377299 -0.053281 -0.039699 -0.054190
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.917195 0.544104 0.930047 0.987077
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.696266 0.739994 1.048117 0.896593
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.293284 0.669478 1.302638 0.481789
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.653293 -28.831598 -2.731161 32.029461
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.301490 -1.364039 -0.114492 1.323518
[Debug] talker-logits-prefill: [3072] first4: -3.051305 -10.105056 -1.549332 -3.043770
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1703 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1047 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=37 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1759 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1163 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=564 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=796 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=783 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1999 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.004818 -0.029755 0.017770 0.080159
[Debug] talker-hidden-step1: [2048] first4: 0.548895 -0.647020 7.054575 -3.797646
[Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1879 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=628 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=1799 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=908 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=132 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=942 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1522 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=317 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=623 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames
[Pipeline] Generated 32 frames
[Pipeline] Generated 40 frames
[Pipeline] Generated 48 frames
[Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] output-audio: [122880] first4: 0.000048 0.000057 0.000046 0.000050
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
[Input] Instruct: male, young adult, moderate pitch [Input] Instruct: male, young adult, moderate pitch
[Input] Lang: english Seed: 42 MaxNewTokens: 64 [Input] Lang: english Seed: 42 MaxNewTokens: 64
@@ -28,3 +106,20 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav [Python] Audio: 120960 samples 24000 Hz 5.04s -> python/tts/tts-python.wav
[Quant] BF16 -> ../models/qwen-talker-1.7b-voicedesign-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf [Quant] BF16 -> ../models/qwen-talker-1.7b-voicedesign-BF16.gguf + ../models/qwen-tokenizer-12hz-BF16.gguf
[GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy [GGML] Cmd: ../build/qwen-tts --model ../models/qwen-talker-1.7b-voicedesign-BF16.gguf --codec ../models/qwen-tokenizer-12hz-BF16.gguf --seed 42 --text qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS for zero shot multilingual text to speech. --instruct male, young adult, moderate pitch --lang english --max-new 64 --dump cpp/tts -o cpp/tts/tts-cpp.wav --greedy
[GGML] Audio: 122880 samples 24000 Hz 5.12s -> cpp/tts/tts-cpp.wav
[Cossim] PromptIDs exact: 100.00% (38 values)
[Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08
[Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] L0 cos: 1.000000 max: 4.0043e-02 mean: 1.2960e-04
[Cossim] L7 cos: 0.999999 max: 9.2588e+00 mean: 1.5759e-03
[Cossim] L14 cos: 0.999999 max: 9.2266e+00 mean: 3.0027e-03
[Cossim] L21 cos: 0.999999 max: 9.2190e+00 mean: 8.4080e-03
[Cossim] L27 cos: 0.999996 max: 2.3640e+01 mean: 4.3644e-02
[Cossim] Final cos: 0.999992 max: 5.2023e-01 mean: 4.8779e-03
[Cossim] Logits cos: 0.999995 max: 6.8526e-02 mean: 9.5837e-03
[Cossim] NextEmbStep0 cos: 1.000000 max: 8.3447e-07 mean: 9.2208e-09
[Cossim] TalkerHiddenStep1 cos: 0.999999 max: 2.2796e-02 mean: 3.3371e-03
[Cossim] CodesFull exact: 19.44% (1008 values)
[Cossim] Audio cos: 0.372872
[Cossim] WAV stft_cos: 0.531607 samples: 120960
tests/tts-Vulkan0-F32.log
+60 -60
View File
@@ -18,15 +18,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72 [Transformer] Loaded: 8 layers, hidden 512, heads 16/16, head_dim 64, FFN 1024, RoPE theta 10000, sliding window 72
[WeightCtx] Loaded 22 tensors, 80.1 MB into backend [WeightCtx] Loaded 22 tensors, 80.1 MB into backend
[Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7 [Upsample] Loaded: 2 blocks (2x ratio per block), channels 1024, dwconv kernel 7
[WeightCtx] Loaded 118 tensors, 199.8 MB into backend [WeightCtx] Loaded 118 tensors, 142.9 MB into backend
[DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 199.8 MB [DAC] Loaded: 4 blocks (strides 8/5/4/3), 24 kHz mono out, weights 142.9 MB
[WeightCtx] Loaded 2 tensors, 6.0 MB into backend [WeightCtx] Loaded 2 tensors, 3.0 MB into backend
[WeightCtx] Loaded 28 tensors, 48.2 MB into backend [WeightCtx] Loaded 28 tensors, 24.1 MB into backend
[SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 48.2 MB [SEANet] Loaded: 4 stages (ratios 4/5/6/8), 64 -> 512 channels, weights 24.1 MB
[WeightCtx] Loaded 96 tensors, 96.1 MB into backend [WeightCtx] Loaded 96 tensors, 96.1 MB into backend
[EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000 [EncTransformer] Loaded: 8 layers, hidden 512, heads 8/8, head_dim 64, FFN 2048, RoPE theta 10000
[WeightCtx] Loaded 1 tensors, 4.0 MB into backend [WeightCtx] Loaded 1 tensors, 2.0 MB into backend
[EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 4.0 MB [EncDownsample] Loaded: k=4 stride=2, 512 -> 512 channels, weights 2.0 MB
[WeightCtx] Loaded 20 tensors, 34.0 MB into backend [WeightCtx] Loaded 20 tensors, 34.0 MB into backend
[EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB [EncQuantizer] Loaded: 16 codebooks (1 semantic + 15 acoustic), 2048 entries x 256 dim, hidden 512, weights 34.0 MB
[Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz [Pipeline] Ready: hop 1920 samples @ 24000 Hz mono, 16 codebooks @ 12.5 Hz
@@ -41,47 +41,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724 [Debug] trailing-text-hidden: [1, 2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724 [Debug] tts-pad-embed: [2048] first4: -0.004532 0.007560 -0.004596 -0.000724
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.377561 -0.053297 -0.039743 -0.054188 [Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.377561 -0.053297 -0.039743 -0.054188
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.920873 0.548539 0.926258 0.994933 [Debug] talker-hidden-prefill-l7: [52, 2048] first4: -0.921009 0.548415 0.926041 0.995244
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.698720 0.743128 1.044047 0.903014 [Debug] talker-hidden-prefill-l14: [52, 2048] first4: -0.699224 0.743413 1.043896 0.903247
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.290339 0.678288 1.299727 0.487553 [Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.290611 0.679320 1.301284 0.486914
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.545020 -29.117104 -2.840297 32.037647 [Debug] talker-hidden-prefill-l27: [52, 2048] first4: -6.566702 -29.101055 -2.837411 32.037216
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.297654 -1.382519 -0.119497 1.328634 [Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.298525 -1.381227 -0.119330 1.328108
[Debug] talker-logits-prefill: [3072] first4: -3.015625 -10.132812 -1.593750 -3.062500 [Debug] talker-logits-prefill: [3072] first4: -3.019531 -10.140625 -1.572266 -3.042969
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1602 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1088 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=1703 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1000 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=1047 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1521 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=37 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1767 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=1759 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1023 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1163 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=635 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=564 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=398 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1820 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=612 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=796 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=157 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=783 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1794 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=1999 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1399 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=1772 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1602.000000 1088.000000 [Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.056980 -0.053437 0.010571 -0.035270 [Debug] next-emb-step0: [2048] first4: 0.004818 -0.029755 0.017770 0.080159
[Debug] talker-hidden-step1: [2048] first4: -1.876469 -3.299715 -0.850600 -0.712180 [Debug] talker-hidden-step1: [2048] first4: 0.534397 -0.656066 7.061725 -3.808899
[Sample] step=1 c0=505 u=-1.0000000000 subseq=16 [Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1752 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=878 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1112 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1879 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=791 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1572 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=628 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1799 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1731 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=279 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1039 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=346 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=908 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=395 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=132 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1033 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=942 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1918 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1522 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1805 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=317 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=193 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=623 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -91,8 +91,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1602.000000 1088.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] output-audio: [122880] first4: 0.000017 0.000020 0.000011 0.000011 [Debug] output-audio: [122880] first4: 0.000047 0.000055 0.000045 0.000047
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08 [Cossim] Embed cos: 1.000000 max: 2.3842e-06 mean: 1.8509e-08
[Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09 [Cossim] TrailingText cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09 [Cossim] TTSPadEmbed cos: 1.000000 max: 8.0466e-07 mean: 9.1670e-09
[Cossim] L0 cos: 0.999998 max: 3.3560e-02 mean: 2.3534e-04 [Cossim] L0 cos: 0.999998 max: 3.6648e-02 mean: 2.3396e-04
[Cossim] L7 cos: 0.999997 max: 2.3657e+01 mean: 2.9852e-03 [Cossim] L7 cos: 0.999998 max: 2.6232e+01 mean: 3.0546e-03
[Cossim] L14 cos: 0.999997 max: 2.3630e+01 mean: 5.2238e-03 [Cossim] L14 cos: 0.999998 max: 2.6172e+01 mean: 5.3008e-03
[Cossim] L21 cos: 0.999997 max: 2.3655e+01 mean: 1.4206e-02 [Cossim] L21 cos: 0.999998 max: 2.6173e+01 mean: 1.3938e-02
[Cossim] L27 cos: 0.999992 max: 2.5899e+01 mean: 8.3456e-02 [Cossim] L27 cos: 0.999994 max: 2.3815e+01 mean: 7.5717e-02
[Cossim] Final cos: 0.999975 max: 6.5904e-01 mean: 9.2044e-03 [Cossim] Final cos: 0.999973 max: 7.9596e-01 mean: 9.0659e-03
[Cossim] Logits cos: 0.999991 max: 8.7137e-02 mean: 1.3245e-02 [Cossim] Logits cos: 0.999991 max: 8.4095e-02 mean: 1.2975e-02
[Cossim] NextEmbStep0 cos: 0.564984 max: 2.6316e-01 mean: 4.8794e-02 [Cossim] NextEmbStep0 cos: 1.000000 max: 8.3447e-07 mean: 9.2208e-09
[Cossim] TalkerHiddenStep1 cos: 0.278505 max: 2.1955e+01 mean: 2.5290e+00 [Cossim] TalkerHiddenStep1 cos: 0.999995 max: 4.9522e-02 mean: 7.1255e-03
[Cossim] CodesFull exact: 0.20% (1008 values) [Cossim] CodesFull exact: 22.42% (1008 values)
[Cossim] Audio cos: 0.000892 [Cossim] Audio cos: 0.387247
[Cossim] WAV stft_cos: 0.049533 samples: 120960 [Cossim] WAV stft_cos: 0.639551 samples: 120960
tests/tts-Vulkan0-Q4_K_M.log
+52 -52
View File
@@ -42,46 +42,46 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] tts-pad-embed: [2048] first4: -0.004540 0.007622 -0.004462 -0.000982 [Debug] tts-pad-embed: [2048] first4: -0.004540 0.007622 -0.004462 -0.000982
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.365522 -0.027829 -0.021236 -0.090642 [Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.365522 -0.027829 -0.021236 -0.090642
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -11.040422 0.707821 5.835257 -2.337338 [Debug] talker-hidden-prefill-l7: [52, 2048] first4: -11.040422 0.707821 5.835257 -2.337338
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -10.607028 0.632244 5.976556 -2.445023 [Debug] talker-hidden-prefill-l14: [52, 2048] first4: -10.607057 0.632322 5.976467 -2.445058
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -9.634210 -0.166934 6.658974 -1.739157 [Debug] talker-hidden-prefill-l21: [52, 2048] first4: -9.635786 -0.167362 6.659793 -1.737383
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -13.693601 -29.542841 7.490469 30.102182 [Debug] talker-hidden-prefill-l27: [52, 2048] first4: -13.698904 -29.517475 7.479723 30.061430
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.651792 -1.468135 0.329833 1.306573 [Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.651991 -1.466754 0.329333 1.304697
[Debug] talker-logits-prefill: [3072] first4: -6.042969 -8.476562 0.566406 -0.104187 [Debug] talker-logits-prefill: [3072] first4: -6.046875 -8.476562 0.581055 -0.080627
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1114 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=295 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=2030 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1894 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=6 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1643 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=2026 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1142 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=359 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=327 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1643 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=387 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=1822 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=1658 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1360 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1809 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=57 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=437 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=841 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=1677 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=732 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=1484 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=77 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=1008 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=278 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=1409 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1114.000000 295.000000 [Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] next-emb-step0: [2048] first4: 0.010285 -0.009700 -0.000257 0.021635 [Debug] next-emb-step0: [2048] first4: 0.040523 -0.070059 0.015252 0.066806
[Debug] talker-hidden-step1: [2048] first4: -1.652356 -4.971395 -0.333058 -0.910279 [Debug] talker-hidden-step1: [2048] first4: 1.290672 -2.405036 4.064333 -2.934217
[Sample] step=1 c0=174 u=-1.0000000000 subseq=16 [Sample] step=1 c0=215 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1924 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=85 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=813 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=681 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=282 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=1385 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=606 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=74 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=672 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1579 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1316 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=739 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=1153 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=1694 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=197 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=1179 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=151 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=243 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=1490 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=823 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=245 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=706 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1707 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=344 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1622 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=586 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=1002 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -91,8 +91,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1114.000000 295.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 2030.000000
[Debug] output-audio: [122880] first4: 0.000016 0.000016 0.000003 0.000016 [Debug] output-audio: [122880] first4: 0.000071 0.000095 0.000097 0.000094
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999680 max: 1.3213e-02 mean: 9.5199e-04 [Cossim] Embed cos: 0.999680 max: 1.3213e-02 mean: 9.5199e-04
[Cossim] TrailingText cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04 [Cossim] TrailingText cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] TTSPadEmbed cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04 [Cossim] TTSPadEmbed cos: 0.999969 max: 2.4252e-03 mean: 2.2081e-04
[Cossim] L0 cos: 0.999207 max: 7.6312e-01 mean: 6.5113e-03 [Cossim] L0 cos: 0.999207 max: 7.6508e-01 mean: 6.5091e-03
[Cossim] L7 cos: 0.997484 max: 1.5604e+02 mean: 4.3659e-01 [Cossim] L7 cos: 0.997484 max: 1.6003e+02 mean: 4.3654e-01
[Cossim] L14 cos: 0.997458 max: 1.5686e+02 mean: 5.4276e-01 [Cossim] L14 cos: 0.997458 max: 1.6086e+02 mean: 5.4263e-01
[Cossim] L21 cos: 0.997066 max: 1.5483e+02 mean: 1.0092e+00 [Cossim] L21 cos: 0.997065 max: 1.5880e+02 mean: 1.0092e+00
[Cossim] L27 cos: 0.988934 max: 5.2996e+02 mean: 4.5230e+00 [Cossim] L27 cos: 0.988971 max: 5.3177e+02 mean: 4.5186e+00
[Cossim] Final cos: 0.935085 max: 2.8574e+01 mean: 5.0807e-01 [Cossim] Final cos: 0.935061 max: 2.8535e+01 mean: 5.0797e-01
[Cossim] Logits cos: 0.889860 max: 9.0592e+00 mean: 1.7072e+00 [Cossim] Logits cos: 0.889198 max: 9.0670e+00 mean: 1.7116e+00
[Cossim] NextEmbStep0 cos: 0.592676 max: 2.6653e-01 mean: 4.5659e-02 [Cossim] NextEmbStep0 cos: 0.832342 max: 1.4996e-01 mean: 3.1096e-02
[Cossim] TalkerHiddenStep1 cos: 0.234535 max: 2.2844e+01 mean: 2.5944e+00 [Cossim] TalkerHiddenStep1 cos: 0.695479 max: 1.1770e+01 mean: 1.7717e+00
[Cossim] CodesFull exact: 0.20% (1008 values) [Cossim] CodesFull exact: 0.69% (1008 values)
[Cossim] Audio cos: -0.007758 [Cossim] Audio cos: -0.001740
[Cossim] WAV stft_cos: 0.067234 samples: 120960 [Cossim] WAV stft_cos: 0.091503 samples: 120960
tests/tts-Vulkan0-Q8_0.log
+53 -53
View File
@@ -41,47 +41,47 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Debug] trailing-text-hidden: [1, 2048] first4: -0.004554 0.007647 -0.004656 -0.000707 [Debug] trailing-text-hidden: [1, 2048] first4: -0.004554 0.007647 -0.004656 -0.000707
[Debug] tts-pad-embed: [2048] first4: -0.004554 0.007647 -0.004656 -0.000707 [Debug] tts-pad-embed: [2048] first4: -0.004554 0.007647 -0.004656 -0.000707
[Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.379852 -0.054862 -0.041081 -0.052085 [Debug] talker-hidden-prefill-l0: [52, 2048] first4: -0.379852 -0.054862 -0.041081 -0.052085
[Debug] talker-hidden-prefill-l7: [52, 2048] first4: -1.402517 0.407066 1.510208 0.573808 [Debug] talker-hidden-prefill-l7: [52, 2048] first4: -1.402396 0.408609 1.507983 0.574799
[Debug] talker-hidden-prefill-l14: [52, 2048] first4: -1.175941 0.612193 1.632609 0.488113 [Debug] talker-hidden-prefill-l14: [52, 2048] first4: -1.175818 0.613162 1.630474 0.489529
[Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.785830 0.492748 1.932347 0.109454 [Debug] talker-hidden-prefill-l21: [52, 2048] first4: -0.786077 0.493968 1.929888 0.109035
[Debug] talker-hidden-prefill-l27: [52, 2048] first4: -7.070812 -29.670967 -2.154939 31.796692 [Debug] talker-hidden-prefill-l27: [52, 2048] first4: -7.063078 -29.640556 -2.152674 31.781174
[Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.319355 -1.399132 -0.090039 1.309577 [Debug] talker-hidden-prefill-final: [52, 2048] first4: -0.318960 -1.397496 -0.089932 1.308749
[Debug] talker-logits-prefill: [3072] first4: -2.992188 -10.062500 -1.494141 -3.066406 [Debug] talker-logits-prefill: [3072] first4: -2.982422 -10.085938 -1.507812 -3.093750
[Sample] step=0 c0=1995 u=-1.0000000000 subseq=0 [Sample] step=0 c0=1995 u=-1.0000000000 subseq=0
[Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1 [Sample-CP] g=0 c=1642 u=-1.0000000000 subseq=1
[Sample-CP] g=1 c=1602 u=-1.0000000000 subseq=2 [Sample-CP] g=1 c=1398 u=-1.0000000000 subseq=2
[Sample-CP] g=2 c=1088 u=-1.0000000000 subseq=3 [Sample-CP] g=2 c=1703 u=-1.0000000000 subseq=3
[Sample-CP] g=3 c=1000 u=-1.0000000000 subseq=4 [Sample-CP] g=3 c=1047 u=-1.0000000000 subseq=4
[Sample-CP] g=4 c=1521 u=-1.0000000000 subseq=5 [Sample-CP] g=4 c=37 u=-1.0000000000 subseq=5
[Sample-CP] g=5 c=1767 u=-1.0000000000 subseq=6 [Sample-CP] g=5 c=1759 u=-1.0000000000 subseq=6
[Sample-CP] g=6 c=1023 u=-1.0000000000 subseq=7 [Sample-CP] g=6 c=1163 u=-1.0000000000 subseq=7
[Sample-CP] g=7 c=635 u=-1.0000000000 subseq=8 [Sample-CP] g=7 c=564 u=-1.0000000000 subseq=8
[Sample-CP] g=8 c=398 u=-1.0000000000 subseq=9 [Sample-CP] g=8 c=1 u=-1.0000000000 subseq=9
[Sample-CP] g=9 c=1820 u=-1.0000000000 subseq=10 [Sample-CP] g=9 c=743 u=-1.0000000000 subseq=10
[Sample-CP] g=10 c=612 u=-1.0000000000 subseq=11 [Sample-CP] g=10 c=796 u=-1.0000000000 subseq=11
[Sample-CP] g=11 c=157 u=-1.0000000000 subseq=12 [Sample-CP] g=11 c=783 u=-1.0000000000 subseq=12
[Sample-CP] g=12 c=1794 u=-1.0000000000 subseq=13 [Sample-CP] g=12 c=1999 u=-1.0000000000 subseq=13
[Sample-CP] g=13 c=1399 u=-1.0000000000 subseq=14 [Sample-CP] g=13 c=82 u=-1.0000000000 subseq=14
[Sample-CP] g=14 c=1772 u=-1.0000000000 subseq=15 [Sample-CP] g=14 c=803 u=-1.0000000000 subseq=15
[Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1602.000000 1088.000000 [Debug] codes-step0: [16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] next-emb-step0: [2048] first4: 0.056649 -0.053523 0.010754 -0.035536 [Debug] next-emb-step0: [2048] first4: 0.004793 -0.029425 0.017809 0.080303
[Debug] talker-hidden-step1: [2048] first4: -1.753798 -3.439198 -0.773368 -0.761658 [Debug] talker-hidden-step1: [2048] first4: 0.538580 -0.664507 7.114338 -3.862589
[Sample] step=1 c0=505 u=-1.0000000000 subseq=16 [Sample] step=1 c0=2042 u=-1.0000000000 subseq=16
[Sample-CP] g=0 c=1752 u=-1.0000000000 subseq=17 [Sample-CP] g=0 c=1230 u=-1.0000000000 subseq=17
[Sample-CP] g=1 c=878 u=-1.0000000000 subseq=18 [Sample-CP] g=1 c=141 u=-1.0000000000 subseq=18
[Sample-CP] g=2 c=1112 u=-1.0000000000 subseq=19 [Sample-CP] g=2 c=1879 u=-1.0000000000 subseq=19
[Sample-CP] g=3 c=791 u=-1.0000000000 subseq=20 [Sample-CP] g=3 c=371 u=-1.0000000000 subseq=20
[Sample-CP] g=4 c=1572 u=-1.0000000000 subseq=21 [Sample-CP] g=4 c=628 u=-1.0000000000 subseq=21
[Sample-CP] g=5 c=267 u=-1.0000000000 subseq=22 [Sample-CP] g=5 c=1799 u=-1.0000000000 subseq=22
[Sample-CP] g=6 c=1731 u=-1.0000000000 subseq=23 [Sample-CP] g=6 c=187 u=-1.0000000000 subseq=23
[Sample-CP] g=7 c=279 u=-1.0000000000 subseq=24 [Sample-CP] g=7 c=645 u=-1.0000000000 subseq=24
[Sample-CP] g=8 c=1039 u=-1.0000000000 subseq=25 [Sample-CP] g=8 c=834 u=-1.0000000000 subseq=25
[Sample-CP] g=9 c=346 u=-1.0000000000 subseq=26 [Sample-CP] g=9 c=908 u=-1.0000000000 subseq=26
[Sample-CP] g=10 c=395 u=-1.0000000000 subseq=27 [Sample-CP] g=10 c=132 u=-1.0000000000 subseq=27
[Sample-CP] g=11 c=1033 u=-1.0000000000 subseq=28 [Sample-CP] g=11 c=942 u=-1.0000000000 subseq=28
[Sample-CP] g=12 c=1918 u=-1.0000000000 subseq=29 [Sample-CP] g=12 c=1522 u=-1.0000000000 subseq=29
[Sample-CP] g=13 c=1805 u=-1.0000000000 subseq=30 [Sample-CP] g=13 c=317 u=-1.0000000000 subseq=30
[Sample-CP] g=14 c=193 u=-1.0000000000 subseq=31 [Sample-CP] g=14 c=623 u=-1.0000000000 subseq=31
[Pipeline] Generated 8 frames [Pipeline] Generated 8 frames
[Pipeline] Generated 16 frames [Pipeline] Generated 16 frames
[Pipeline] Generated 24 frames [Pipeline] Generated 24 frames
@@ -91,8 +91,8 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Pipeline] Generated 56 frames [Pipeline] Generated 56 frames
[Pipeline] Generated 64 frames [Pipeline] Generated 64 frames
[Pipeline] Generation done : 64 frames [Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1602.000000 1088.000000 [Debug] codes-full: [64, 16] first4: 1995.000000 1642.000000 1398.000000 1703.000000
[Debug] output-audio: [122880] first4: 0.000018 0.000022 0.000014 0.000015 [Debug] output-audio: [122880] first4: 0.000047 0.000055 0.000042 0.000045
[WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16 [WAV] Wrote cpp/tts/tts-cpp.wav: 122880 samples, 24000 Hz, mono S16
[Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav [Pipeline] Wrote 122880 samples (5.12 s) -> cpp/tts/tts-cpp.wav
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo... [Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -111,15 +111,15 @@ load_backend: loaded CPU backend from /mnt/workspace/qwentts.cpp/build/libggml-c
[Cossim] Embed cos: 0.999989 max: 2.2227e-03 mean: 1.6673e-04 [Cossim] Embed cos: 0.999989 max: 2.2227e-03 mean: 1.6673e-04
[Cossim] TrailingText cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05 [Cossim] TrailingText cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05
[Cossim] TTSPadEmbed cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05 [Cossim] TTSPadEmbed cos: 0.999999 max: 1.8711e-04 mean: 4.2146e-05
[Cossim] L0 cos: 0.999992 max: 4.5462e-02 mean: 6.5980e-04 [Cossim] L0 cos: 0.999992 max: 4.9368e-02 mean: 6.5849e-04
[Cossim] L7 cos: 0.999986 max: 2.6321e+01 mean: 3.2151e-02 [Cossim] L7 cos: 0.999986 max: 1.8340e+01 mean: 3.2028e-02
[Cossim] L14 cos: 0.999986 max: 2.6197e+01 mean: 4.0514e-02 [Cossim] L14 cos: 0.999986 max: 1.8247e+01 mean: 4.0458e-02
[Cossim] L21 cos: 0.999983 max: 2.6375e+01 mean: 7.7104e-02 [Cossim] L21 cos: 0.999983 max: 1.8454e+01 mean: 7.6089e-02
[Cossim] L27 cos: 0.999945 max: 6.2261e+01 mean: 3.3700e-01 [Cossim] L27 cos: 0.999947 max: 6.4114e+01 mean: 3.3413e-01
[Cossim] Final cos: 0.999666 max: 1.4233e+00 mean: 3.7771e-02 [Cossim] Final cos: 0.999670 max: 1.4440e+00 mean: 3.7552e-02
[Cossim] Logits cos: 0.999756 max: 3.9106e-01 mean: 6.4052e-02 [Cossim] Logits cos: 0.999767 max: 4.0209e-01 mean: 6.3427e-02
[Cossim] NextEmbStep0 cos: 0.565000 max: 2.6307e-01 mean: 4.8794e-02 [Cossim] NextEmbStep0 cos: 0.999990 max: 1.0252e-03 mean: 2.4056e-04
[Cossim] TalkerHiddenStep1 cos: 0.277044 max: 2.2239e+01 mean: 2.5255e+00 [Cossim] TalkerHiddenStep1 cos: 0.999732 max: 2.9859e-01 mean: 5.0384e-02
[Cossim] CodesFull exact: 0.20% (1008 values) [Cossim] CodesFull exact: 8.23% (1008 values)
[Cossim] Audio cos: 0.006878 [Cossim] Audio cos: 0.217182
[Cossim] WAV stft_cos: 0.064563 samples: 120960 [Cossim] WAV stft_cos: 0.489313 samples: 120960