 Qiaolin Yuandshuwenn
|
e068355831
|
[spec decoding] supports step 0 in adaptive spec decoding (updating draft kv cache without draft decoding) (#23994)
Co-authored-by: shuwenn <2508695655@qq.com>
|
2026-06-15 22:21:26 -07:00 |
|
|
Thomas Wang
|
800aaefc9e
|
[AMD] Annotate ATOM source for imported v4 unified attention kernels (#28392)
|
2026-06-15 22:12:12 -07:00 |
|
|
iridiumine
|
486ec150d0
|
[NPU] Add NPU fallback for fused Triton gating kernels (#28293)
|
2026-06-16 11:37:05 +08:00 |
|
|
Shu Wang
|
32685874f3
|
Reenable MNNVL backend for FlashInfer allreduce fusion (#23402)
|
2026-06-15 20:19:15 -07:00 |
|
|
shuwenn
|
b23477af44
|
bench: infer tokenizer from serving model info (#28195)
|
2026-06-15 20:03:12 -07:00 |
|
|
zhangxiaolei
|
063ab89ac1
|
DeepSeek-V4 Online Compress support MTP (#26471)
|
2026-06-15 19:56:07 -07:00 |
|
|
huangtingwei
|
b5bcd76a41
|
[HiCache & JIT Kernel] Refactoring HiCache Write-Back Kernel (#21631)
|
2026-06-15 19:44:27 -07:00 |
|
|
Mick
|
a4a8a614b1
|
[diffusion] UX: suppress noisy diffusers torchao warning (#28317)
|
2026-06-16 10:32:21 +08:00 |
|
|
Mick
|
01e45762ba
|
[diffusion] feat: use srt custom allreduce for tp groups (#28324)
|
2026-06-16 09:22:47 +08:00 |
|
|
YAMY
|
b3be2e7402
|
[dsv4] Pad MLA decode q-heads to 64 (not full n_heads) for FlashMLA head64 kernel (#27954)
|
2026-06-15 17:18:10 -07:00 |
|
|
Liangsheng Yin
|
14f6348524
|
[Fix] Demote OpenAIServingResponses init failure log to one-line WARNING (#28349)
|
2026-06-15 16:58:33 -07:00 |
|
|
Hanming Lu
|
81dcb00673
|
[Spec v2] Use decode kernel for TRT-LLM MHA draft extend (#28241)
|
2026-06-15 16:57:08 -07:00 |
|
|
Kangyan-ZhouandClaude Fable 5
|
cad43d3212
|
[CI] Reclaim leaked /dev/shm segments on server startup (#28089)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
|
2026-06-15 16:14:08 -07:00 |
|
|
Jia Guo
|
4ed698a491
|
fix(fa3): no NaN embeddings with fa_skip_kv_cache under piecewise CUDA graph (#27343)
|
2026-06-15 13:46:22 -07:00 |
|
|
YAMY
|
f870bf1ed0
|
[dsv4] Prewarm MHC prenorm kernel at startup (#27986)
|
2026-06-15 13:26:26 -07:00 |
|
|
Lianmin ZhengandIan O'Connell
|
7e629a2f8c
|
Allow overriding tokenizer path in benchmark harness (#28280)
Co-authored-by: Ian O'Connell <ianoc@meta.com>
|
2026-06-15 13:07:50 -07:00 |
|
|
|
33719cfb31
|
[PD] Optimize SWA allocation (#28085)
Co-authored-by: cctry <cctry@fb.com>
Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
|
2026-06-15 11:01:55 -07:00 |
|
|
Shangming Cai
|
378e66d248
|
[PD] Remove outdated backend whitelist for decode radix cache (#28238)
Signed-off-by: Shangming Cai <csmthu@gmail.com>
|
2026-06-15 22:36:05 +08:00 |
|
|
Trevor Morris
|
20f4272109
|
fix: Fix DSR1 perf regression due to unnecessarily falling back to triton gemm (#28073)
|
2026-06-15 09:45:09 -04:00 |
|
|
YAMY
|
d5899b95c4
|
fix(qwen3.5): keep CUDA dual-stream overlap (regressed by #25885) (#27868)
|
2026-06-15 09:44:21 -04:00 |
|
|
littleyellowbicycle
|
09e9c4fde3
|
【bugfix】The NPU's forward_dsa_prepare_npu also needs special handling for is_nextn (#28118)
|
2026-06-15 21:31:39 +08:00 |
|
|
Kurkur
|
e985422b2b
|
[Fix][MTP][MM] Fix EAGLE v2 chunked-prefill next-token chain crash on multimodal models due to placeholder tokens (#27863)
|
2026-06-15 21:30:45 +08:00 |
|
|
Mick
|
818808d152
|
[diffusion] optimize: optimize causal conv3d vae padding (#28204)
|
2026-06-15 20:18:08 +08:00 |
|
|
iridiumine
|
3df6e2f968
|
[NPU] Add MiMo-V2-Flash manual testcases (#28223)
|
2026-06-15 19:57:49 +08:00 |
|
|
|
eb349efb14
|
[EPD][BugFix] Fix encode_with_global_cache_mooncake (#28031)
Co-authored-by: 晟海 <huangtingwei.htw@antgroup.com>
Co-authored-by: Michael Qiu <qiudayu.qdy@antgroup.com>
|
2026-06-15 19:47:52 +08:00 |
|
|
|
c4ec39a785
|
[AMD] refactor sparse MLA decode kernel for Deepseek V4 triton backend (#28265)
Co-authored-by: Raiden-Makoto <Raiden-Makoto@users.noreply.github.com>
Co-authored-by: yichiche@amd.com <jacky.cheng>
|
2026-06-15 03:26:59 -07:00 |
|
|
Thomas Wang
|
da12f36629
|
[AMD] Refactor unified_kv attention metadata to data class and fuse c4/128 out_loc (#28275)
|
2026-06-15 02:59:01 -07:00 |
|
|
Lianmin Zhengandgemini-code-assist[bot]
|
3b419f66da
|
[JIT] Track angle-bracket includes in source hash (#28273)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
|
2026-06-15 02:58:14 -07:00 |
|
|
Zhonghua Deng
|
2a33724c9b
|
[perf] Reuse a pooled HTTP session for multimodal URL downloads (#28056)
Signed-off-by: Abatom <abzhonghua@gmail.com>
|
2026-06-15 17:47:47 +08:00 |
|
|
McZyWu
|
bf186cf8fc
|
bugfix revise interface get cpu copy for npu mem pool to align with gpu (#27802)
|
2026-06-15 17:24:10 +08:00 |
|
|
Duyi-Wang
|
63df86f5e7
|
[AMD] Skip eplb bookkeeping and topk remap when EPLB is not in use on mori-ep / HIP (#22985) (#28188)
|
2026-06-14 23:00:23 -07:00 |
|
|
Mick
|
578e936d8d
|
[diffusion] feat: persist torch.compile inductor/triton cache across restarts (#28205)
|
2026-06-15 13:34:19 +08:00 |
|
|
|
07b9108348
|
[Diffusion] FLUX: fuse FeedForward GELU into up-proj GEMM (cublasLt epilogue) (#28166)
Co-authored-by: BBuf <xiaoyu.zhang@radixark.net>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
|
2026-06-15 13:00:26 +08:00 |
|
|
Prajjandprajjwal1
|
441b75ee69
|
[quantization] NVFP4 MoE: split fused w13 gate/up global scales (#27588)
Co-authored-by: prajjwal1 <prajjwal1@protonmail.com>
|
2026-06-14 21:18:36 -07:00 |
|
|
Ryan Zzzandzhujunyu
|
ce9fad7196
|
[Bugfix][DeepSeek-V4] Fix Spec V2 Draft Input ID Dtype for DP Collectives (#28043)
Co-authored-by: zhujunyu <zhujunyu.666@bytedance.com>
|
2026-06-14 21:11:53 -07:00 |
|
|
weireweire
|
bf38a0b03d
|
Fix disaggregated decode load token accounting (#25736)
|
2026-06-15 11:37:41 +08:00 |
|
|
|
0417951a86
|
[Bug Fix] Validate tokenizer-dependent features with skip_tokenizer_init (#27882)
Co-authored-by: Randall <randall@iterationlab.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
|
2026-06-14 20:07:52 -07:00 |
|
|
Zyann
|
37505eca27
|
feat: report multimodal (image/audio/video) token counts in usage.prompt_tokens_details (#27122)
|
2026-06-15 11:04:10 +08:00 |
|
|
Mohammad Miadh Angkad
|
69b02ea68a
|
[Distributed] Guard torch symm mem all-reduce sizes (#24548)
|
2026-06-14 18:28:57 -07:00 |
|
|
Xinyuan Tong
|
1a66059c4e
|
[Spec] Restore index_share_for_mtp_iteration in EAGLE V2 draft worker (#28192)
|
2026-06-14 18:01:11 -07:00 |
|
|
Lianmin Zheng
|
f18d38d040
|
Revert "[AMD][Quantization] Online MXFP4 quantization 2/N - FP8 to MXFP4 requantization on AMD GPUs" (#28213)
|
2026-06-14 13:34:06 -07:00 |
|
|
Mick
|
3cb29f6747
|
[diffusion] feat: use regional torch.compile (compile_repeated_blocks) for DiT of diffusers backend (#28193)
|
2026-06-15 00:34:10 +08:00 |
|
|
Mick
|
ec36dde580
|
[diffusion] feat: add --warmup-mode enum server arg (#28184)
|
2026-06-14 23:09:04 +08:00 |
|
|
Mick
|
582bd23f71
|
[diffusion] feat: enable spatial-shard vae decode across GPUs (#28071)
|
2026-06-14 20:19:44 +08:00 |
|
|
|
d72314808f
|
[JIT Kernel] Multi-GPU test/bench framework for custom all-reduce + TP QKNorm (#26706)
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: ziyi.xu <ziyi.xu@radixark.ai>
|
2026-06-14 17:20:36 +08:00 |
|
|
Mick
|
5331de0f8c
|
[diffusion] chore: resolve model_index.json Hub-first with local-cache fallback (#28177)
|
2026-06-14 16:48:38 +08:00 |
|
|
Humphrey
|
8c334e2224
|
fix(io_struct): index extra_key per sub-request in batched GenerateReqInput (#26971)
|
2026-06-14 00:50:38 -07:00 |
|
|
Mick
|
1456eb612d
|
[diffusion] CI: tighten perf baselines (#28123)
|
2026-06-14 15:50:21 +08:00 |
|
|
Jimmy Shongandgithub-actions[bot]
|
54acffc864
|
Eval accuracy gpqa aime25 mixins (#27102)
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
|
2026-06-14 00:49:12 -07:00 |
|
|
Qiaolin Yu
|
f293ddf3ce
|
[perf] reduce overhead of fill_ids list reconstruction and decref (#27965)
|
2026-06-14 00:41:11 -07:00 |
|