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

logs

Browse Source
This commit is contained in:
Pascal
2026-07-05 10:25:31 +02:00
parent 2700d4c746
commit c39ef15bab
12 changed files with 105 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/clone-CPU-BF16.log
+10 -10
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -33,7 +33,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Audio-Resample] Done: 380672 -> 414337 samples
[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
[Pipeline] Speaker encoder lazy loaded in 6 ms
[Pipeline] Speaker encoder lazy loaded in 5 ms
[Debug] mel-spk: [1618, 128] first4: -6.330339 -6.170074 -6.011195 -6.243161
[Debug] mel-hann: [1024] first4: 0.000000 0.000009 0.000038 0.000085
[Debug] mel-basis: [128, 513] first4: 0.000000 0.033550 0.008569 0.000000
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 54 ms
[Pipeline] Codec encoder lazy loaded in 45 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.112500 13.760122 -0.329032 2.993861
[Debug] enc-transformer-out: [430, 512] first4: 0.526661 3.208130 -0.164395 0.605249
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000529 0.000353 -0.000865 -0.000453
[Perf] PromptBuild 2.8 ms
[Perf] Prefill 380.0 ms (T_ctx prefill)
[Perf] TTFA 408.1 ms (first frame codes)
[Perf] TalkerDecode 2920.2 ms (64 frames, 45.63 ms/frame)
[Perf] CodePredictor 1452.0 ms (22.69 ms/frame)
[Perf] PromptBuild 2.3 ms
[Perf] Prefill 380.6 ms (T_ctx prefill)
[Perf] TTFA 408.5 ms (first frame codes)
[Perf] TalkerDecode 2916.5 ms (64 frames, 45.57 ms/frame)
[Perf] CodePredictor 1464.6 ms (22.88 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 1021.6 ms
[Perf] Total 5778.3 ms (64 frames, 68.32 ms/frame AR, audio 5.12 s, RTF 1.129)
[Perf] CodecDecode 1034.6 ms
[Perf] Total 5800.0 ms (64 frames, 68.46 ms/frame AR, audio 5.12 s, RTF 1.133)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-CPU-F32.log
+10 -10
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -33,7 +33,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Audio-Resample] Done: 380672 -> 414337 samples
[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
[Pipeline] Speaker encoder lazy loaded in 6 ms
[Pipeline] Speaker encoder lazy loaded in 8 ms
[Debug] mel-spk: [1618, 128] first4: -6.330339 -6.170074 -6.011195 -6.243161
[Debug] mel-hann: [1024] first4: 0.000000 0.000009 0.000038 0.000085
[Debug] mel-basis: [128, 513] first4: 0.000000 0.033550 0.008569 0.000000
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 34 ms
[Pipeline] Codec encoder lazy loaded in 58 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.120292 13.742489 -0.348104 3.009387
[Debug] enc-transformer-out: [430, 512] first4: 0.526076 3.125370 -0.189241 0.630764
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000450 0.000327 -0.000890 -0.000492
[Perf] PromptBuild 3.8 ms
[Perf] Prefill 623.1 ms (T_ctx prefill)
[Perf] TTFA 690.7 ms (first frame codes)
[Perf] TalkerDecode 5586.7 ms (64 frames, 87.29 ms/frame)
[Perf] CodePredictor 3937.5 ms (61.52 ms/frame)
[Perf] PromptBuild 3.9 ms
[Perf] Prefill 632.5 ms (T_ctx prefill)
[Perf] TTFA 700.2 ms (first frame codes)
[Perf] TalkerDecode 5539.4 ms (64 frames, 86.55 ms/frame)
[Perf] CodePredictor 3887.4 ms (60.74 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 1012.2 ms
[Perf] Total 11164.9 ms (64 frames, 148.82 ms/frame AR, audio 5.12 s, RTF 2.181)
[Perf] CodecDecode 1033.3 ms
[Perf] Total 11098.2 ms (64 frames, 147.30 ms/frame AR, audio 5.12 s, RTF 2.168)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-CPU-Q4_K_M.log
+9 -9
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 8 ms
[Pipeline] Codec encoder lazy loaded in 7 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.120292 13.742489 -0.348104 3.009387
[Debug] enc-transformer-out: [430, 512] first4: 0.867046 3.016766 0.059672 0.395116
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 417.000000
[Debug] output-audio: [122880] first4: -0.000160 -0.000284 -0.001051 -0.000894
[Perf] PromptBuild 2.8 ms
[Perf] Prefill 414.8 ms (T_ctx prefill)
[Perf] TTFA 425.6 ms (first frame codes)
[Perf] TalkerDecode 990.4 ms (64 frames, 15.48 ms/frame)
[Perf] CodePredictor 415.2 ms (6.49 ms/frame)
[Perf] PromptBuild 3.3 ms
[Perf] Prefill 412.6 ms (T_ctx prefill)
[Perf] TTFA 424.0 ms (first frame codes)
[Perf] TalkerDecode 990.1 ms (64 frames, 15.47 ms/frame)
[Perf] CodePredictor 413.0 ms (6.45 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 1031.5 ms
[Perf] Total 2856.3 ms (64 frames, 21.96 ms/frame AR, audio 5.12 s, RTF 0.558)
[Perf] CodecDecode 1015.1 ms
[Perf] Total 2835.6 ms (64 frames, 21.93 ms/frame AR, audio 5.12 s, RTF 0.554)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-CPU-Q8_0.log
+11 -11
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 9 ms
[Pipeline] Codec encoder lazy loaded in 8 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.120292 13.742489 -0.348104 3.009387
[Debug] enc-transformer-out: [430, 512] first4: 0.716872 2.959070 -0.221450 0.852651
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000527 0.000324 -0.000983 -0.000676
[Perf] PromptBuild 3.3 ms
[Perf] Prefill 615.3 ms (T_ctx prefill)
[Perf] TTFA 631.6 ms (first frame codes)
[Perf] TalkerDecode 1675.4 ms (64 frames, 26.18 ms/frame)
[Perf] CodePredictor 650.8 ms (10.17 ms/frame)
[Perf] PromptBuild 2.9 ms
[Perf] Prefill 612.7 ms (T_ctx prefill)
[Perf] TTFA 628.6 ms (first frame codes)
[Perf] TalkerDecode 1690.1 ms (64 frames, 26.41 ms/frame)
[Perf] CodePredictor 658.2 ms (10.28 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 993.5 ms
[Perf] Total 3948.8 ms (64 frames, 36.35 ms/frame AR, audio 5.12 s, RTF 0.771)
[Perf] CodecDecode 1021.4 ms
[Perf] Total 3996.0 ms (64 frames, 36.69 ms/frame AR, audio 5.12 s, RTF 0.780)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
@@ -180,5 +180,5 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] CodesFull exact: 2.68% (1008 values)
[Cossim] Audio cos: -0.002547
[Cossim] WAV stft_cos: 0.174666 samples: 120960
[Cossim] Audio cos: -0.012062
[Cossim] WAV stft_cos: 0.175184 samples: 120960
tests/clone-CUDA0-BF16.log
+10 -10
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -33,7 +33,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Audio-Resample] Done: 380672 -> 414337 samples
[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
[Pipeline] Speaker encoder lazy loaded in 3 ms
[Pipeline] Speaker encoder lazy loaded in 4 ms
[Debug] mel-spk: [1618, 128] first4: -6.330339 -6.170074 -6.011195 -6.243161
[Debug] mel-hann: [1024] first4: 0.000000 0.000009 0.000038 0.000085
[Debug] mel-basis: [128, 513] first4: 0.000000 0.033550 0.008569 0.000000
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 63 ms
[Pipeline] Codec encoder lazy loaded in 42 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.111631 13.770183 -0.302467 2.984188
[Debug] enc-transformer-out: [430, 512] first4: 0.531195 3.238480 -0.146394 0.612285
@@ -140,14 +140,14 @@ ggml_backend_cuda_graph_compute: CUDA graph warmup complete
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000465 0.000259 -0.001065 -0.000736
[Perf] PromptBuild 1.2 ms
[Perf] Prefill 13.3 ms (T_ctx prefill)
[Perf] TTFA 23.1 ms (first frame codes)
[Perf] TalkerDecode 188.4 ms (64 frames, 2.94 ms/frame)
[Perf] CodePredictor 303.3 ms (4.74 ms/frame)
[Perf] PromptBuild 1.3 ms
[Perf] Prefill 13.5 ms (T_ctx prefill)
[Perf] TTFA 23.3 ms (first frame codes)
[Perf] TalkerDecode 189.6 ms (64 frames, 2.96 ms/frame)
[Perf] CodePredictor 304.1 ms (4.75 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 14.6 ms
[Perf] Total 521.9 ms (64 frames, 7.68 ms/frame AR, audio 5.12 s, RTF 0.102)
[Perf] CodecDecode 14.7 ms
[Perf] Total 524.3 ms (64 frames, 7.72 ms/frame AR, audio 5.12 s, RTF 0.102)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-CUDA0-F32.log
+7 -7
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -140,14 +140,14 @@ ggml_backend_cuda_graph_compute: CUDA graph warmup complete
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000739 0.000593 -0.000747 -0.000435
[Perf] PromptBuild 1.4 ms
[Perf] PromptBuild 1.3 ms
[Perf] Prefill 27.4 ms (T_ctx prefill)
[Perf] TTFA 37.5 ms (first frame codes)
[Perf] TalkerDecode 300.4 ms (64 frames, 4.69 ms/frame)
[Perf] CodePredictor 388.9 ms (6.08 ms/frame)
[Perf] TTFA 37.4 ms (first frame codes)
[Perf] TalkerDecode 304.7 ms (64 frames, 4.76 ms/frame)
[Perf] CodePredictor 389.8 ms (6.09 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 14.9 ms
[Perf] Total 734.2 ms (64 frames, 10.77 ms/frame AR, audio 5.12 s, RTF 0.143)
[Perf] CodecDecode 14.8 ms
[Perf] Total 739.3 ms (64 frames, 10.85 ms/frame AR, audio 5.12 s, RTF 0.144)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-CUDA0-Q4_K_M.log
+10 -10
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -33,7 +33,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Audio-Resample] Done: 380672 -> 414337 samples
[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
[Pipeline] Speaker encoder lazy loaded in 3 ms
[Pipeline] Speaker encoder lazy loaded in 4 ms
[Debug] mel-spk: [1618, 128] first4: -6.330339 -6.170074 -6.011195 -6.243161
[Debug] mel-hann: [1024] first4: 0.000000 0.000009 0.000038 0.000085
[Debug] mel-basis: [128, 513] first4: 0.000000 0.033550 0.008569 0.000000
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 7 ms
[Pipeline] Codec encoder lazy loaded in 8 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.134336 13.731120 -0.351051 3.013485
[Debug] enc-transformer-out: [430, 512] first4: 1.140990 3.713432 0.131578 0.616251
@@ -182,14 +182,14 @@ ggml_backend_cuda_graph_compute: CUDA graph warmup complete
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1042.000000 390.000000 881.000000
[Debug] output-audio: [122880] first4: 0.000087 0.000212 -0.000383 -0.000346
[Perf] PromptBuild 55.2 ms
[Perf] Prefill 14.4 ms (T_ctx prefill)
[Perf] TTFA 78.1 ms (first frame codes)
[Perf] TalkerDecode 220.3 ms (64 frames, 3.44 ms/frame)
[Perf] CodePredictor 327.3 ms (5.11 ms/frame)
[Perf] PromptBuild 56.4 ms
[Perf] Prefill 14.6 ms (T_ctx prefill)
[Perf] TTFA 79.2 ms (first frame codes)
[Perf] TalkerDecode 224.9 ms (64 frames, 3.51 ms/frame)
[Perf] CodePredictor 338.0 ms (5.28 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 15.7 ms
[Perf] Total 634.2 ms (64 frames, 8.56 ms/frame AR, audio 5.12 s, RTF 0.124)
[Perf] CodecDecode 23.8 ms
[Perf] Total 658.8 ms (64 frames, 8.80 ms/frame AR, audio 5.12 s, RTF 0.129)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-CUDA0-Q8_0.log
+3 -3
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -143,8 +143,8 @@ ggml_backend_cuda_graph_compute: CUDA graph warmup complete
[Perf] PromptBuild 1.5 ms
[Perf] Prefill 15.2 ms (T_ctx prefill)
[Perf] TTFA 24.0 ms (first frame codes)
[Perf] TalkerDecode 153.4 ms (64 frames, 2.40 ms/frame)
[Perf] CodePredictor 245.1 ms (3.83 ms/frame)
[Perf] TalkerDecode 152.3 ms (64 frames, 2.38 ms/frame)
[Perf] CodePredictor 246.1 ms (3.85 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 15.0 ms
[Perf] Total 431.5 ms (64 frames, 6.23 ms/frame AR, audio 5.12 s, RTF 0.084)
tests/clone-Vulkan0-BF16.log
+10 -10
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 44 ms
[Pipeline] Codec encoder lazy loaded in 45 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.110593 13.765666 -0.317055 2.981106
[Debug] enc-transformer-out: [430, 512] first4: 0.536260 3.195749 -0.156796 0.610111
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000515 0.000357 -0.000915 -0.000575
[Perf] PromptBuild 1.9 ms
[Perf] Prefill 31.2 ms (T_ctx prefill)
[Perf] TTFA 41.1 ms (first frame codes)
[Perf] TalkerDecode 247.4 ms (64 frames, 3.87 ms/frame)
[Perf] CodePredictor 418.8 ms (6.54 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 35.3 ms
[Perf] Total 735.7 ms (64 frames, 10.41 ms/frame AR, audio 5.12 s, RTF 0.144)
[Perf] PromptBuild 2.0 ms
[Perf] Prefill 31.7 ms (T_ctx prefill)
[Perf] TTFA 41.7 ms (first frame codes)
[Perf] TalkerDecode 248.0 ms (64 frames, 3.88 ms/frame)
[Perf] CodePredictor 415.5 ms (6.49 ms/frame)
[Perf] HostCompose 0.2 ms (c0 sample + next emb)
[Perf] CodecDecode 30.8 ms
[Perf] Total 729.2 ms (64 frames, 10.37 ms/frame AR, audio 5.12 s, RTF 0.142)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-Vulkan0-F32.log
+9 -9
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -51,7 +51,7 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[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] Codec encoder lazy loaded in 34 ms
[Pipeline] Codec encoder lazy loaded in 35 ms
[Debug] audio-input: [412800] first4: -0.000546 -0.001162 -0.000871 0.000340
[Debug] seanet-out: [430, 512] first4: -0.138303 13.699992 -0.351357 3.025051
[Debug] enc-transformer-out: [430, 512] first4: 0.526569 3.116264 -0.195330 0.625080
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000673 0.000478 -0.000828 -0.000482
[Perf] PromptBuild 2.3 ms
[Perf] Prefill 19.7 ms (T_ctx prefill)
[Perf] TTFA 31.0 ms (first frame codes)
[Perf] TalkerDecode 367.9 ms (64 frames, 5.75 ms/frame)
[Perf] CodePredictor 504.1 ms (7.88 ms/frame)
[Perf] PromptBuild 9.1 ms
[Perf] Prefill 19.6 ms (T_ctx prefill)
[Perf] TTFA 37.9 ms (first frame codes)
[Perf] TalkerDecode 369.2 ms (64 frames, 5.77 ms/frame)
[Perf] CodePredictor 512.4 ms (8.01 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 28.8 ms
[Perf] Total 924.0 ms (64 frames, 13.63 ms/frame AR, audio 5.12 s, RTF 0.180)
[Perf] CodecDecode 28.9 ms
[Perf] Total 940.4 ms (64 frames, 13.78 ms/frame AR, audio 5.12 s, RTF 0.184)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-Vulkan0-Q4_K_M.log
+8 -8
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 417.000000
[Debug] output-audio: [122880] first4: -0.000101 0.000003 -0.000779 -0.000457
[Perf] PromptBuild 2.6 ms
[Perf] Prefill 16.6 ms (T_ctx prefill)
[Perf] TTFA 25.6 ms (first frame codes)
[Perf] TalkerDecode 175.7 ms (64 frames, 2.75 ms/frame)
[Perf] CodePredictor 328.4 ms (5.13 ms/frame)
[Perf] PromptBuild 2.8 ms
[Perf] Prefill 16.8 ms (T_ctx prefill)
[Perf] TTFA 26.1 ms (first frame codes)
[Perf] TalkerDecode 181.9 ms (64 frames, 2.84 ms/frame)
[Perf] CodePredictor 328.5 ms (5.13 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 31.6 ms
[Perf] Total 556.1 ms (64 frames, 7.88 ms/frame AR, audio 5.12 s, RTF 0.109)
[Perf] CodecDecode 29.7 ms
[Perf] Total 561.0 ms (64 frames, 7.98 ms/frame AR, audio 5.12 s, RTF 0.110)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...
tests/clone-Vulkan0-Q8_0.log
+8 -8
View File
@@ -1,4 +1,4 @@
[Qwen] qwentts.cpp f84fc05 (2026-07-04)
[Qwen] qwentts.cpp bcac463 (2026-07-05)
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/git/qwentts.cpp/build/libggml-cuda.so
@@ -122,14 +122,14 @@ load_backend: loaded CPU backend from /mnt/workspace/git/qwentts.cpp/build/libgg
[Pipeline] Generation done : 64 frames
[Debug] codes-full: [64, 16] first4: 27.000000 1408.000000 1824.000000 400.000000
[Debug] output-audio: [122880] first4: 0.000658 0.000394 -0.000943 -0.000566
[Perf] PromptBuild 2.6 ms
[Perf] Prefill 16.7 ms (T_ctx prefill)
[Perf] TTFA 25.8 ms (first frame codes)
[Perf] TalkerDecode 199.6 ms (64 frames, 3.12 ms/frame)
[Perf] CodePredictor 331.5 ms (5.18 ms/frame)
[Perf] PromptBuild 2.7 ms
[Perf] Prefill 17.1 ms (T_ctx prefill)
[Perf] TTFA 26.5 ms (first frame codes)
[Perf] TalkerDecode 201.3 ms (64 frames, 3.14 ms/frame)
[Perf] CodePredictor 334.3 ms (5.22 ms/frame)
[Perf] HostCompose 0.1 ms (c0 sample + next emb)
[Perf] CodecDecode 30.1 ms
[Perf] Total 581.8 ms (64 frames, 8.30 ms/frame AR, audio 5.12 s, RTF 0.114)
[Perf] CodecDecode 30.3 ms
[Perf] Total 586.9 ms (64 frames, 8.37 ms/frame AR, audio 5.12 s, RTF 0.115)
[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
[Input] Prompt: 100 chars: qwentts.cpp is a minimal C++17 and GGML port of Qwen3-TTS fo...