Sy03
d530938b5b
[Model] Reuse frame-local KV in MOSS-TTS Local
...
Cache each position once while generating the audio codebooks, preserving frame-local RoPE positions and the full-prefix reference path. Share compiled batch and position shapes to avoid specialization-limit fallback. Add one compact parity regression covering frame reuse.
Signed-off-by: Sy03 <1370724210@qq.com >
2026-09-08 06:37:00 +08:00
Sy03
30d28a9a04
[Test] Consolidate MOSS-TTS batching regression coverage
...
Combine overlapping row/snapshot checks, cover masked and valid attention rows in one batch, reuse the real codec session in grouping tests, and move output-policy coverage into the existing entrypoint module. Preserve both CUDA synchronization regressions and reduce the added suite from 29 to 15 cases.
Signed-off-by: Sy03 <1370724210@qq.com >
2026-09-08 03:03:32 +08:00
Sy03
55855fee9c
[Model] Reduce MOSS-TTS Local intermediate output overhead
...
Keep internal talker control outputs FINAL_ONLY while codec audio remains incremental. Reapply pipeline sampling constraints after default streaming coercion, log routine coercions at debug level, and cover explicit/default parameters with regression tests. Document matched high-concurrency statistics settings.
Signed-off-by: Sy03 <1370724210@qq.com >
2026-09-08 01:15:46 +08:00
Sy03
251ef4f48f
[Model] Optimize MOSS-TTS Local batched execution and streaming codec
...
Signed-off-by: Sy03 <1370724210@qq.com >
2026-09-07 16:20:49 +08:00
BruceLoveDecimal and Claude Opus 5
1e74807c35
[Feature][Moss-TTS] Reference Audio Preprocessing ( #4982 )
...
Signed-off-by: BruceLoveDecimal <39156883+BruceLoveDecimal@users.noreply.github.com >
Signed-off-by: liuqihao <liuqihao970610@gmail.com >
Co-authored-by: Claude Opus 5 <noreply@anthropic.com >
2026-09-04 16:36:26 -07:00
37c11e6e4f
[Perf][Qwen3-TTS] Fuse QKV and gate_up projections in code predictor ( #5791 )
...
Signed-off-by: sheenlin <sheenlin@tencent.com >
Signed-off-by: Gao Han <hgaoaf@connect.ust.hk >
Co-authored-by: sheenlin <sheenlin@tencent.com >
Co-authored-by: yurain <yurain26@users.noreply.github.com >
Co-authored-by: Gao Han <hgaoaf@connect.ust.hk >
2026-08-20 19:22:40 +00:00
Sy03
26e003715c
[Perf] Enable async schedule for MOSS-TTS ( #6241 )
...
Signed-off-by: Sy03 <1370724210@qq.com >
2026-08-19 06:36:19 +00:00
Gao Han
78c144f3a8
[CI failed] Revert "[Perf][Qwen3-TTS] Fuse QKV and gate_up projections in code predictor" ( #5777 )
2026-08-04 15:21:38 +00:00
6fdc4ca62d
[Perf][Qwen3-TTS] Fuse QKV and gate_up projections in code predictor ( #4958 )
...
Signed-off-by: sheenlin <sheenlin@tencent.com >
Co-authored-by: sheenlin <sheenlin@tencent.com >
Co-authored-by: yurain <yurain26@users.noreply.github.com >
2026-08-04 11:59:32 +00:00
wangyu
7c17111ff5
[Refactor][CI] Reorganize test tree, and move slow e2e to weekly ( #4951 )
...
Signed-off-by: wangyu <410167048@qq.com >
2026-07-29 23:53:03 +08:00
Zhou Taichang
a560ed184d
[Rebase] Rebase to vllm v0.24.0 ( #4709 )
2026-06-30 19:41:06 +08:00
yangyonggit and Yueqian Lin
1228a109f0
[Bugfix] MOSS-TTS: fix cross-request audio corruption under batching ( #4415 )
...
Signed-off-by: leo.yang <leo.yang.engineer@gmail.com >
Co-authored-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com >
2026-06-18 18:11:51 +00:00
akshatvishu
d8a72be206
fix(models): remove invalid sampling_metadata parameter from logits_processor ( #4456 )
...
Signed-off-by: akshatvishu <akshatnayak197@gmail.com >
2026-06-16 14:41:40 +08:00
yangyonggit
a3834e32cf
feat(moss-tts): add CUDA Graph support for codec decoder ( #4157 )
...
Signed-off-by: leo.yang <leo.yang.engineer@gmail.com >
2026-06-11 22:17:31 +00:00
Sy03