Jan Bernlöhr
f3ecc3688f
Fix Gemma3 ModelOpt kv-scale loading ( #25794 )
2026-06-09 17:53:21 -07:00
Muqi Li and zqlcode
365b7dab9a
fix(schema): update tokens_after_end ( #27017 )
...
Co-authored-by: zqlcode <1309223143@qq.com >
2026-06-09 16:52:35 -07:00
Mohammad Miadh Angkad
bc82086ef8
Remove FlashInfer GB transport workaround ( #27453 )
...
Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com >
2026-06-09 16:48:49 -07:00
Wenqi and wenqi
98fe7e326e
fix(gemma4): register image/video/audio token_regex for HF-expanded prompts ( #26320 )
...
Co-authored-by: wenqi <wenqi@convergence.ai >
2026-06-09 16:30:13 -07:00
Yueming Yuan
53a4b51f8c
Fix GLM NextN draft value head dim ( #26049 )
2026-06-09 16:13:37 -07:00
David Wang
4455abd164
dflash piecewise cuda graphs support ( #27468 )
2026-06-09 15:44:19 -07:00
decb88e0e3
Support spec v2 for Frozen-KV MTP; remove v1 worker ( #27607 )
...
Co-authored-by: Khoa Pham <khoa.pham@radixark.ai >
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com >
2026-06-09 15:30:20 -07:00
7f730edfdc
fix: correct off-by-one in vocab boundary check for token validation ( #22367 )
...
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com >
Co-authored-by: fzyzcjy <ch271828n@outlook.com >
2026-06-09 15:24:39 -07:00
weizhoublue
fde4004429
[Fix] Reset positions tensor in CUDA graph runner when batch size differs from captured size ( #24401 )
...
Signed-off-by: weizhoublue <weizhou.lan@daocloud.io >
2026-06-09 15:03:58 -07:00
jacky.cheng
2fef951fe8
[AMD] Replace fp8 mla with fp8 mha kernel for diffusion model aiter backend ( #23927 )
2026-06-09 14:46:41 -07:00
ziang663
42322947aa
[BUG FIX]Fix DSA CPU offload mamba indices signature ( #27645 )
2026-06-09 14:03:38 -07:00
Lianmin Zheng and lmzheng
eb8dceda44
Defer DeepGEMM PDL setup to worker init ( #27671 )
...
Co-authored-by: lmzheng <lmzheng@fb.com >
2026-06-09 13:52:30 -07:00
Liangsheng Yin
186f1e300a
[CI] Move JIT kernel tests + benchmarks to test/registered/jit; add in-package guard ( #27644 )
2026-06-09 12:37:39 -07:00
Bi Xue
8ae328e5f0
[sgl] Fix kimi-k2.5 EAGLE3 MLA draft embeds for batched MM prefill ( #27647 )
2026-06-09 11:26:48 -07:00
Michael
5babb902a9
[AMD] fix: handle per-frame 4D shift in native scale-shift kernel ( #27581 )
2026-06-09 10:31:00 -07:00
Xiaoyu Zhang and Claude Opus 4.8
aa18a68ac5
[diffusion] Run LTX-2 VAE decode in channels_last_3d (faster decode, lower peak memory) ( #27431 )
...
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com >
2026-06-09 23:26:40 +08:00
fzyzcjy
1368717248
Add more testing for chunked prefill ( #27506 )
2026-06-09 20:19:30 +08:00
fzyzcjy
609f5f549c
Add mixed-prefix gsm8k eval and its CPU unit test ( #27502 )
2026-06-09 20:17:41 +08:00
fdcd28a08d
[NPU] Enable consistency checking for diffusion tests ( #27283 )
...
Co-authored-by: Elizaveta Martirosian <elizaveta.martirosian@gmail.com >
Co-authored-by: Elizaveta Martirosian <you@example.com >
2026-06-09 15:01:45 +03:00
McZyWu
c6be251c5b
[NPU] RL update_weights_from_disk/ tensor /distributed ( #26717 )
2026-06-09 16:52:36 +08:00
AndyLi429
cd6efcb947
[NPU][Bugfix] fix MTP accuracy regression on Qwen3 hybrid models ( #27202 )
2026-06-09 15:47:23 +08:00
2218622f50
Fix spec v2 stop output boundary ( #25980 )
...
Co-authored-by: gss <2783977641@qq.com >
Co-authored-by: hnyls2002 <lsyincs@gmail.com >
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com >
2026-06-09 00:32:21 -07:00
Zaili Wang
d981b7b9c4
[Fix] Avoid applying cuda graph input-buffer registry on non-cuda devices ( #27549 )
2026-06-09 00:24:55 -07:00
Mohammad Miadh Angkad
a287ab83c0
Fix Gemma4 NVFP4 MoE default attention backend ( #26791 )
2026-06-09 14:33:08 +08:00
d145a6127a
fix: stop-string check misses early matches during speculative decoding ( #23802 )
...
Co-authored-by: xythink <xythink@users.noreply.github.com >
Co-authored-by: hnyls2002 <lsyincs@gmail.com >
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com >
2026-06-08 23:25:58 -07:00
LJL36 and Zhangheng
991689fd0d
fix(hiradix): wait for extra pool IO ( #27550 )
...
Co-authored-by: Zhangheng <hzh0425@apache.org >
2026-06-09 14:17:36 +08:00
Erik Wijmans
dff695c76b
[lora] Exclude finished requests from running_loras ( #27597 )
2026-06-08 23:05:43 -07:00
Khoa Pham and Cursor
9a3e845fc1
[Spec] Add nvtx to spec regions ( #27615 )
...
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-06-08 22:58:03 -07:00
5c0b2859e8
[diffusion] rl: extract post-training weight apis into mixins and add tensor update/checker paths ( #22817 )
...
Co-authored-by: dreamyang-liu <nikolaliu@icloud.com >
Co-authored-by: Xiaole Guo <vera0315@connect.hku.hk >
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com >
Co-authored-by: Xiaole Guo <119391905+gxlvera@users.noreply.github.com >
Co-authored-by: zhihengy <zhihengy@andrew.cmu.edu >
Co-authored-by: MikukuOvO <MikukuOvO@users.noreply.github.com >
2026-06-09 13:57:47 +08:00
Liangsheng Yin
cae022aa88
[JIT] Reuse JIT kernel build cache across CI runs ( #27605 )
2026-06-08 22:14:48 -07:00
Michael
a32aeb688a
[AMD] Fix AttributeError in GeneratedSharedPrefixDataset.from_args for in-process callers ( #27580 )
2026-06-08 21:53:14 -07:00
c2eae96c56
MSCCL++ Integration ( #22734 )
...
Co-authored-by: Caio Rocha <caiorocha@microsof.com >
Co-authored-by: empyreus <rjsouza1995@gmail.com >
2026-06-08 21:13:13 -07:00
Jan Bernlöhr and XueLiang Yang
15c801f726
fix(server): clamp piecewise_cuda_graph_max_tokens to context_length ( #22516 )
...
Co-authored-by: XueLiang Yang <102161631+xueliangyang-oeuler@users.noreply.github.com >
2026-06-08 19:43:54 -07:00
jianan-gu and Ma Mingfei
db143e5212
[Intel GPU][Encoder] Add xpu_attn backend for encoder vision attention ( #26460 )
...
Co-authored-by: Ma Mingfei <mingfei.ma@intel.com >
2026-06-09 09:47:44 +08:00
ea66b2cca7
[XPU] Enable NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 on Intel XPU backend ( #24390 )
...
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com >
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Yao Matrix <matrix.yao@intel.com >
2026-06-09 09:46:12 +08:00
Cao E
71e8258783
Improve registration in cpu_graph_runner ( #26635 )
2026-06-09 09:35:33 +08:00
Polisetty V R K Jyothendra Varma
d7c8b9ab9f
[Intel GPU] Enable fused_experts in fp8.py for quantized models on XPU ( #27533 )
...
Signed-off-by: P V R K Jyothendra Varma <polisetty.v.r.k.jyothendra.varma@intel.com >
2026-06-09 09:22:30 +08:00
Estrella-xx
6abef38627
[NPU]adaptation to support deterministic inference ( #21197 )
2026-06-09 09:22:09 +08:00
Jianhong Zhang
ab70153b62
[XPU][NIXL] Use uint64 for XPU address arithmetic in prep handle builders ( #27415 )
2026-06-09 09:18:36 +08:00
Liangsheng Yin
3fe6bc390b
[Spec] Naming cleanup: contiguous draft-loc kernel + accepted->accept ( #27599 )
2026-06-08 15:04:58 -07:00
Khoa Pham
c95179bc85
[Spec] Fuse small kenrels under gather_spec_extras ( #27233 )
2026-06-08 15:02:04 -07:00
Liangsheng Yin
b5c64b94d5
[Spec] Rename token resolver to _resolve_spec_v2_tokens; remove dead V1 helpers ( #27552 )
2026-06-08 14:42:21 -07:00
YAMY and Yuwei An
ca66e6fb5e
[BCG] Support breakable CUDA graph for DeepSeek V4 DP attention ( #25195 )
...
Co-authored-by: Yuwei An <ayw.sirius19@gmail.com >
2026-06-08 13:54:58 -07:00
Mohammad Miadh Angkad
dc24a26821
Fix GPT-OSS MXFP4 hidden size reshape on SM10X ( #27528 )
...
Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com >
2026-06-08 13:26:54 -07:00
Liangsheng Yin
28c1a3cb45
[Spec] Deprecate Spec V1 ( #25464 )
2026-06-08 13:10:27 -07:00
ea1d190ed0
[ROCm] dsv4: remove the redundant fp8 scale transpose-copy on decode ( #27289 )
...
Co-authored-by: Zhiyao Jiang <jessicajiang324@gmail.com >
Co-authored-by: Thomas Wang <thomawan@amd.com >
2026-06-08 11:49:34 -07:00
Lianmin Zheng
fca4ef9d69
Fix SWA pool resolution for EAGLE draft workers ( #27491 )
2026-06-08 11:00:29 -07:00
Lianmin Zheng
bcb5645629
Fix TRTLLM target verify query metadata ( #27495 )
2026-06-08 10:58:48 -07:00
32bedbf88e
[diffusion] model: support SANA-WM with streaming support ( #27531 )
...
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com >
Co-authored-by: sjmshsh <88866917+sjmshsh@users.noreply.github.com >
Co-authored-by: Mick <mickjagger19@icloud.com >
2026-06-09 01:10:14 +08:00
MARATRIX
61e4132bc2
[MUSA] bump torchada version to 0.1.59 and workaround PCG limitation. ( #27537 )
...
Signed-off-by: yafeng.li <yafeng.li@mthreads.com >
2026-06-08 08:45:39 -07:00
Jan Bernlöhr