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

Commit Graph

  • 7942e61b3a feat: anti-loop guards + KV cache configurabile master enne2 2026-08-18 15:10:02 +02:00
  • 18742ea656 fix: volumi compose relativi alla root del repo (../../models, ../../voices) enne2 2026-08-13 22:20:51 +02:00
  • 8b9361803e docker: containerizzazione tts-server con GPU Vulkan (Dockerfile multi-stage + compose + entrypoint voci .spk/.rvq) enne2 2026-08-13 20:06:55 +02:00
  • 3be42af9bb tts-server: normalizza i numeri interi 5+ cifre in parole italiane (Qwen3-TTS li leggeva cifra per cifra) enne2 2026-08-13 19:54:50 +02:00
  • a8a7716b53 build: split CUDA Docker images into cuda12 (12.9.x) / cuda13 (13.3.x), fix Blackwell Ultra arch gate (#23)
    Stefan GeraschGitHubGary
         2026-08-07 11:26:45 +02:00
  • 37d62e42a0 ci: lowercase the GHCR image repository path (#22)
    Stefan GeraschGitHubGary
         2026-08-07 08:27:14 +02:00
  • 07f837494a cmake: emit Pascal (61-real) only on CUDA 12.x toolkits Pascal 2026-08-07 08:03:54 +02:00
  • 978fc08f5a Add Dockerfile (cpu/cuda/vulkan), entrypoint, docs, and a GHCR build/release workflow (#15)
    Stefan GeraschGitHubGary
         2026-08-07 07:52:22 +02:00
  • 7b6ed4f6db codec: drop the fused streaming tail Pascal 2026-08-05 18:20:08 +02:00
  • abab6b3bf3 tests: refresh the clone grid logs Pascal 2026-07-30 23:05:37 +02:00
  • 9a9a425260 tests: exercise the fused decode in every cossim harness Pascal 2026-07-30 22:54:44 +02:00
  • 26dd8adbf0 predictor: unroll the frame into one cgraph and sample in standard ops Pascal 2026-07-30 22:53:58 +02:00
  • 35ebe5376b nit: drop obsolete ABI comments Pascal 2026-07-25 19:06:43 +02:00
  • 13686e8d34 logs Pascal 2026-07-25 18:56:39 +02:00
  • d03ffb97f9 api: derived codec left context, chunk width hoisted to qt_init Pascal 2026-07-25 18:56:28 +02:00
  • 710a52af75 ./format.sh (clang-format) Pascal 2026-07-25 11:26:59 +02:00
  • fb9d097946 Merge pull request #14 from DerGary/codec-chunk-dur-server-flags PascalandGitHub 2026-07-25 11:25:34 +02:00
  • 144a79f5cc tts-server: expose --codec-chunk-dur / --codec-left-dur Gary 2026-07-23 09:11:10 +02:00
  • 82cd05b9f3 ggml: fork update Pascal 2026-07-21 19:34:30 +02:00
  • ba4c7f7838 logs Pascal 2026-07-20 20:36:48 +02:00
  • 521503f50b vulkan: support misaligned get_rows operands Pascal 2026-07-20 20:06:02 +02:00
  • 877d8e1389 logs Pascal 2026-07-20 19:27:39 +02:00
  • 53c84fe1be engine: batched multi lane codec streaming, shared lockstep chunk ramp Pascal 2026-07-20 19:13:05 +02:00
  • 9d7768e8ee ggml: fix conv_1d result layout for batched input Pascal 2026-07-20 19:12:51 +02:00
  • b2e36056eb graphs: express the SwiGLU MLPs through ggml_swiglu_split Pascal 2026-07-20 16:50:28 +02:00
  • 37ea692be6 engine: true parallel batching of the talker and predictor, per slot codec streams Pascal 2026-07-20 14:32:14 +02:00
  • fbff3317a4 ggml: bump submodule for full CUDA GET_ROWS type coverage Pascal 2026-07-19 22:30:22 +02:00
  • e93292bee1 ggml: bump submodule for CUDA k-quant GET_ROWS support Pascal 2026-07-19 21:29:06 +02:00
  • 7c4fea29ce ggml: fork update Pascal 2026-07-19 20:30:30 +02:00
  • 95b4840ad3 loader: validate raw u32 counts before the (int) cast Pascal 2026-07-17 21:46:24 +02:00
  • d5b05efff7 nit: realign endpoint tables on the widened voice routes Pascal 2026-07-17 08:29:40 +02:00
  • 0d1b25b863 clang-format Pascal 2026-07-17 08:20:35 +02:00
  • 8d556c9806 server: compatible with OpenAI API, add model alias karlandPascal 2026-07-09 13:18:24 +08:00
  • 19bbf94dcc pipeline-codec: reset sched before encode alloc RaduAdumitroaeiandPascal 2026-07-12 15:23:47 +02:00
  • 0e2b672329 backend: drop the process-wide backend cache inherited from acestep.cpp Pascal 2026-07-17 07:49:13 +02:00
  • 7bb91886f6 ggml: fork update Pascal 2026-07-14 15:14:03 +02:00
  • d17c33d4ee tts: reuse decode scratch buffers across streaming steps Pascal 2026-07-12 09:25:38 +02:00
  • bb250f57e4 backend: dedup consecutive identical ggml log lines Pascal 2026-07-09 15:28:07 +02:00
  • e93b3fedd5 ggml: fork update Pascal 2026-07-08 19:51:08 +02:00
  • 31e1eb648e ggml: fork update Pascal 2026-07-08 19:23:44 +02:00
  • 0725d2e53b codec: adaptive chunk width on the streaming decode Pascal 2026-07-06 22:43:52 +02:00
  • 206ad06fde talker: static decode graphs per attention window class Pascal 2026-07-06 22:26:19 +02:00
  • 2862be4152 predictor: static graphs replayed with a single id upload per step Pascal 2026-07-06 21:46:08 +02:00
  • 66ebf32b2b ggml: fork update Pascal 2026-07-06 15:32:34 +02:00
  • 73fe0c67bb server: sampling overrides on the speech endpoint Pascal 2026-07-05 15:01:37 +02:00
  • 1a680e8816 codec: primed state snapshot LRU keyed by the reference content Pascal 2026-07-05 14:34:27 +02:00
  • 7a4c799f9c server: reject unknown voice names on base models Pascal 2026-07-05 13:45:49 +02:00
  • 0f1c8572c7 qwen: lazy speaker encoder load in the voice ref extraction path Pascal 2026-07-05 13:45:38 +02:00
  • 62dec12580 server: cloned voice registry over the OpenAI surface Pascal 2026-07-05 12:28:55 +02:00
  • a37ff074ff codec: guard the KV ring against models with a wider sliding window Pascal 2026-07-05 10:36:58 +02:00
  • c39ef15bab logs Pascal 2026-07-05 10:25:31 +02:00
  • 2700d4c746 codec: stateful frame by frame streaming decode on a static graph Pascal 2026-07-05 10:25:21 +02:00
  • bcac46352e logs Pascal 2026-07-05 07:47:07 +02:00
  • 415ef56330 tts: static decode graphs and device resident hidden bridge Pascal 2026-07-05 07:43:09 +02:00
  • e2e381f4f0 codec: persistent graph arena for the streaming decode path Pascal 2026-07-05 07:42:50 +02:00
  • f84fc05292 logs Pascal 2026-07-04 22:47:51 +02:00
  • 4057d0331f tts: seed the vocoder left context with the ICL reference tail Pascal 2026-07-04 22:47:41 +02:00
  • 3e3b6c1712 tts: lazy encoder loading and in graph embedding gathers Pascal 2026-07-04 21:35:03 +02:00
  • c9d9fa1c6b ggml: fork update Pascal 2026-07-04 21:19:51 +02:00
  • 3ee7bdd8c8 tts: persistent graph arenas and padded attention windows Pascal 2026-07-04 14:14:59 +02:00
  • 46c99d5889 cmake: default GGML_CUDA_GRAPHS to ON Pascal 2026-07-04 13:45:08 +02:00
  • 0b4ef05d4c ggml: fork update Pascal 2026-07-04 10:55:26 +02:00
  • 3676f0dc41 doc Pascal 2026-07-04 01:54:23 +02:00
  • 3a3069b8fd utf8: normalize inbound text bytes to UTF-8 Pascal 2026-07-04 01:14:00 +02:00
  • 09d3bc90ab ggml: fork update Pascal 2026-07-03 22:38:45 +02:00
  • 9dbe7ea26a tests: absolute paths are prohibited Pascal 2026-06-24 19:00:33 +02:00
  • c7fb781937 Merge pull request #5 from andimarafioti/codex/voice-ref-abi PascalandGitHub 2026-06-24 17:33:58 +02:00
  • 5ce30d34a4 Add voice reference extraction ABI Andres Marafioti 2026-06-24 14:47:09 +02:00
  • 4536dcdce2 cleaning Pascal 2026-06-21 17:30:21 +02:00
  • 26fcea5468 ggml: fork update Pascal 2026-06-19 16:01:29 +02:00
  • 0bf4a18b22 codec: add pre-encoded voice reference (--ref-spk / --ref-rvq) Pascal 2026-06-11 21:45:13 +02:00
  • e8e33629c1 graph: reset scheduler before remaining shared-sched graph allocs Pascal 2026-06-09 19:41:39 +02:00
  • f79b23a567 Merge pull request #4 from andimarafioti/reset-scheduler-before-talker-alloc PascalandGitHub 2026-06-09 12:01:12 +02:00
  • d862d0e10a Reset scheduler before talker graph allocation Andres Marafioti 2026-06-09 11:44:59 +02:00
  • df66c67706 tts: add --stream-by-line, one utterance and one WAV header per line PascalandJeffrey van Binsbergen 2026-06-06 23:13:46 +02:00
  • ed8052eeb3 clang-format Pascal 2026-06-06 22:46:56 +02:00
  • ca0c779f49 tts: default language to auto, NULL lang selects auto, reject NULL text Pascal 2026-06-06 13:31:51 +02:00
  • f3cfa5cf47 server: add OpenAI compatible TTS server (chunked PCM streaming, WAV one-shot) Pascal 2026-06-06 12:47:10 +02:00
  • eda8b59092 docs Pascal 2026-05-31 19:20:57 +02:00
  • 8aba0f012a logs Pascal 2026-05-31 17:49:31 +02:00
  • 08b79d1209 prompt: cut TTFA by projecting text and ICL codec embeds on the backend Pascal 2026-05-31 17:44:06 +02:00
  • 199a65813a logs Pascal 2026-05-31 15:41:34 +02:00
  • 5442c2f84c perf: add per-stage timer instrumentation to qwentts and omnivoice Pascal 2026-05-31 15:05:48 +02:00
  • 127bde4eea docs Pascal 2026-05-31 14:09:28 +02:00
  • 1c62494faa docs Pascal 2026-05-31 13:45:25 +02:00
  • 29a7fa0f97 test logs Pascal 2026-05-31 02:39:03 +02:00
  • a62fde62e6 prompt: drop dead host projection fallback, qt_throw on backend failure Pascal 2026-05-30 16:44:16 +02:00
  • cea3260b64 clang-format Pascal 2026-05-30 16:15:31 +02:00
  • 300bb09944 Merge pull request #3 from andimarafioti/optimize-prompt-ttfa PascalandGitHub 2026-05-30 16:13:12 +02:00
  • 207da32e75 nits Pascal 2026-05-30 16:10:10 +02:00
  • a0f504ce4c Optimize prompt TTFA Andres Marafioti 2026-05-30 15:41:29 +02:00
  • b7d601ff66 nits Pascal 2026-05-20 14:15:09 +02:00
  • 85ad801950 Merge pull request #1 from smashingtags/master PascalandGitHub 2026-05-20 14:10:11 +02:00
  • dc8d2d59a5 tests: add Intel Arc A310 SYCL generation log smashingtags 2026-05-19 19:40:48 -04:00
  • 8608504a5e cmake: add Intel Arc GPU support via SYCL backend smashingtags 2026-05-19 16:15:49 -04:00
  • 55b7234d21 cmake: scope /utf-8 to C and C++ so nvcc does not treat it as an input file Pascal 2026-05-17 14:41:39 +02:00
  • 8e1df7329a ggml: sync Pascal 2026-05-17 14:19:59 +02:00
  • 50b97b12fc ggml: address 0cc4m review for Vulkan fused snake activation Pascal 2026-05-17 11:38:29 +02:00
  • 5523143942 tests Pascal 2026-05-15 00:02:37 +02:00
  • 4bf578ec7b qwen-tts: drop --text, align help on omnivoice style Pascal 2026-05-14 23:48:27 +02:00