biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity
14,997 Commits 2 Branches 140 Tags
ef2c4f778df5aa07a44e663330e2dfdc16927d2a
Commit Graph

606 Commits

Author SHA1 Message Date
Elvir Crnčević
ef2c4f778d [Bugfix] Zero-init MLA attention output buffers to prevent NaN from CUDA graph padding (#37442) 
Signed-off-by: Elvir Crncevic <elvircrn@gmail.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-19 00:28:37 +00:00
Andy Lo
577df69b26 [Bugfix] Fix KV scales inconsistency in fp8 MLA & FlashInfer kv_cache_dtype "auto" leading to gibberish (#37054) 
Signed-off-by: Andy Lo <andy@mistral.ai>
 2026-03-18 23:07:29 +00:00
Divakar Verma
e6c4797704 [ROCm][Quantization] add fp8xfp8 attn support for rocm_aiter_unified_attn (#36927) 
Signed-off-by: Divakar Verma <divakar.verma@amd.com>
 2026-03-18 08:49:32 +08:00
Andrey Talman
68f783a727 [Torch 2.11] Guard torch._C._cpu attribute checks for forward compatibility (#35673) 
Signed-off-by: atalman <atalman@fb.com>
 2026-03-17 18:47:59 +00:00
Benjamin Chislett
8a680463fa [Bugfix] Fix NemotronH MTP + Chunked Prefill (#35447) 2026-03-17 07:07:33 +01:00
Vadim Gimpelson
6c1cfbad32 Support non-contiguous KV cache in TRTLLM fp8 dequant kernel (#36867) 
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
Signed-off-by: Vadim Gimpelson <156319763+vadiklyutiy@users.noreply.github.com>
Co-authored-by: Pavani Majety <pavanimajety@gmail.com>
 2026-03-16 17:48:42 -07:00
Matthew Bonanni
93f3c8e531 [Misc] Add float16 to CacheDType (#37199) 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-16 13:24:48 -07:00
Matthew Bonanni
c88ea8338b [MTP][Sparse MLA] Take advantage of native MTP support in indexer when possible (#36982) 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-16 13:51:21 -04:00
haosdent
ca1954d58c [Bugfix] Disable cross-layer KV cache for MLA attention backends (#37090) 
Signed-off-by: haosdent <haosdent@gmail.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
 2026-03-16 19:03:10 +02:00
Itay Etelis
5ae685c1c8 [Bugfix] Relax TRTLLM KV cache contiguity assertion for cross-layer layout (#34158) 
Signed-off-by: Itay Etelis <itay.etelis@ibm.com>
Co-authored-by: Itay Etelis <itay.etelis@ibm.com>
 2026-03-16 11:20:51 -04:00
haosdent
116ed130f4 [Bugfix] Fix GDN attention crash with mixed decode/spec-decode batches (#34871) 
Signed-off-by: haosdent <haosdent@gmail.com>
 2026-03-16 10:30:23 +01:00
Vadim Gimpelson
8374387bd8 [FlashInfer] Revert block_size 16 + head_size 256 workaround on Blackwell (#36987) 
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
 2026-03-16 09:04:29 +00:00
Dimitrios Bariamis
367cf5cd3e [Feat][Bugfix] Enable additional dimension for Flashinfer MLA and fix routing dtype (#36931) 
Signed-off-by: Dimitrios Bariamis <12195802+dbari@users.noreply.github.com>
Co-authored-by: Dimitrios Bariamis <12195802+dbari@users.noreply.github.com>
 2026-03-13 16:41:16 -07:00
Rohan Potdar
a4ad9db541 Enable RoPE+KV cache fusion for ROCm AITER FA (non-shuffle layout) (#35786) 
Signed-off-by: Rohan138 <rohanpotdar138@gmail.com>
 2026-03-13 07:33:22 +00:00
Matthew Bonanni
f444c05c32 [Attention] Use FA4 for MLA prefill (#34732) 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-12 12:10:17 -04:00
grimulkan
a1257fd1ea [Kernel] Add FP8 KV cache support to Triton MLA decode attention (#34597) 
Signed-off-by: grimulkan <grimulkan@gmail.com>
 2026-03-12 08:32:34 -07:00
Wuxun Zhang
e584dce52b Add XPU MLA Sparse backend for DeepSeek v3.2 (#33230) 
Signed-off-by: Zhang, Wuxun <wuxun.zhang@intel.com>
 2026-03-11 19:19:15 +08:00
JartX
a40ee486f2 [Bugfix] Add Multiple of 16 block_size to triton fallback on rocm Attention to support qwen3_5 (#35923) 
Signed-off-by: JartX <sagformas@epdcenter.es>
Co-authored-by: akaratza <akaratza@amd.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
 2026-03-11 07:45:57 +00:00
pschlan-amd
eac2dc2b41 AITER MLA backend: Avoid CPU sync in _build_decode (#35765) 
Signed-off-by: Patrick Schlangen <pschlan@amd.com>
 2026-03-11 07:25:00 +00:00
Matthew Bonanni
5f77ef15ae [Misc][Attention] Clean up unused method in CPU_ATTN (#36673) 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-10 21:27:22 -07:00
Benjamin Chislett
9040cd40af [DSV3.2][MTP] Optimize Indexer MTP handling (#36723) 
Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>
 2026-03-11 12:16:56 +08:00
Woosuk Kwon
195d1ca3e8 [Minor] Enhance error message for TRTLLM decode uniformity check (#36609) 
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
 2026-03-10 15:38:45 -07:00
Pleaplusone
82f3f30e26 [ROCm][Perf] Enable sparse_mla's cudagraph on ROCm platform (#35719) 
Signed-off-by: ganyi <ygan@amd.com>
 2026-03-10 09:14:35 -07:00
Matthew Bonanni
9095cbbfb6 [Bugfix][Sparse MLA] report indexer CG support properly (#36519) 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-10 09:14:31 -07:00
Vadim Gimpelson
4ff8c3c8f9 [BUGFIX][Mamba][Qwen3.5] Zero freed SSM cache blocks on GPU (#35219) 
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
 2026-03-10 03:32:20 -07:00
Woosuk Kwon
006aea17d7 [BugFix] Remove incorrect assert in split_decodes_and_prefills (#36553) 
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
 2026-03-09 20:02:02 -07:00
Andreas Karatzas
c174d54f86 [ROCm][CI] Fix ROCm attention backend validation for head sizes, block sizes, and compute capability checks (#36292) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
 2026-03-09 12:02:41 -05:00
Roberto L. Castro
580864d81e [Attention][Perf][Kernel] Replace torch.cat with vectorized CUDA kernel MLA query concat - DeepSeek-V3.2 (#34917) 
Signed-off-by: LopezCastroRoberto <rocastro@redhat.com>
Signed-off-by: Roberto L. Castro <38211239+LopezCastroRoberto@users.noreply.github.com>
 2026-03-09 09:50:36 -07:00
Matthew Bonanni
77a73458e3 Reapply [Attention] Refactor check_and_update_config (#35122) 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-09 07:17:14 -07:00
cong-or
747431044d feat(attention): extract KV-cache update from FlexAttention backend (#36263) 
Signed-off-by: cong-or <conchubhar.gannon@gmail.com>
 2026-03-08 20:40:12 -07:00
Wei Zhao
379689d533 [Perf] Support FP8 KV cache for Flashinfer MLA Sparse (#35891) 2026-03-07 13:51:54 -08:00
Mengtao (Martin) Yuan
1a9718085c Fix CUDA graph decode capture crash in AITER FlashAttention (#36042) 
Signed-off-by: Martin Yuan <myuan@meta.com>
Co-authored-by: Martin Yuan <myuan@meta.com>
 2026-03-06 18:12:07 -08:00
Chuan (Richard) Li
c188749bcd [ROCm] Support MLA with nhead<16 and FP8 KV cache for TP=8 (Kimi K2.5/Linear) (#35850) 
Signed-off-by: Li <chuali@amd.com>
 2026-03-06 20:24:03 +00:00
Andreas Karatzas
807d680337 [ROCm][CI] Fix tool use test stability - disable skinny GEMM, prefix caching, eliminate batch variance (#35553) 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
 2026-03-06 15:15:12 +08:00
Rohan Potdar
c5362c739f Reenable features for ROCm attention backends (#36185) 
Signed-off-by: Rohan138 <rohanpotdar138@gmail.com>
 2026-03-05 20:21:06 -08:00
Dor Huri
ebed80a7c8 [Performance] Extract KV-cache update from TreeAttention backend (#35384) 
Signed-off-by: dorhuri123 <dor.huri1@live.biu.ac.il>
 2026-03-06 00:22:43 +00:00
Frank Wang
a57c877f18 [BugFix] Fallback from FA4->FA2 for Batch Invariance (#36059) 
Signed-off-by: frankwang28 <frank.wbb@hotmail.com>
 2026-03-05 14:05:56 -05:00
Jiayi Yan
6a895197fa [Bugfix][CI] fix typos (#34934) 
Signed-off-by: 1195343015 <1195343015@qq.com>
Signed-off-by: Jiayi Yan <66017932+1195343015@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
 2026-03-05 17:05:46 +00:00
Sage Moore
8c760b6ab6 [ROCm] Refactor ROCm attention backend selection logic (#35246) 
Signed-off-by: Sage Moore <sage@neuralmagic.com>
 2026-03-05 10:51:26 -06:00
Harry Mellor
17dc9c7fc9 [CI] Bump mypy version (#34950) 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
 2026-03-04 20:55:11 +00:00
sungsoo ha
6cb901093f [Core] Add All-to-All communication backend for DCP (#34883) 
Signed-off-by: Sungsoo Ha <sungsooh@nvidia.com>
Signed-off-by: sungsoo ha <hasungsoo@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
 2026-03-04 10:01:57 -05:00
Lucas Wilkinson
28ef9ba399 [BugFix] Add support for MTP num_speculative_tokens > 1 with sparse MLA (#34552) 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
 2026-03-03 07:21:57 -08:00
ElizaWszola
d9c7730877 [Performance] Extract kv update ops from MLA attention backends (#34627) 
Signed-off-by: ElizaWszola <ewszola@redhat.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Di Wu <dw2761@nyu.edu>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
 2026-03-02 10:43:19 -05:00
wangxiyuan
510bc9e1df [Misc] Cleanup useless current_platform import (#35715) 
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
 2026-03-02 09:36:54 +00:00
Lucas Wilkinson
8b5014d3dd [Attention] FA4 integration (#32974) 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
 2026-03-01 23:44:57 +00:00
zhanqiuhu
57a96e26c9 Revert "[Bugfix] Disable TRTLLM attention with KV transfer enabled (#33192)" (#34832) 
Signed-off-by: Zhanqiu Hu <zh338@cornell.edu>
 2026-03-01 22:32:37 +00:00
Asaf Gardin
bbf81f9a92 [Mamba1] - Kernel Level Chunk Alignment for Prefix Caching (#34798) 
Signed-off-by: Josephasafg <ajgard7@gmail.com>
 2026-03-01 20:40:23 +08:00
Chauncey
7e08c22b8c [Feat] Add CUDA torch fallbacks for fp8_mqa_logits/fp8_paged_mqa_logits_torch function (#35271) 
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
 2026-02-28 10:12:00 +00:00
Micah Williamson
0edf101d2b [ROCm] Add stablelm Head Size 80 To Supported Head Sizes For ROCM_ATTN (#35527) 
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
 2026-02-28 12:16:34 +08:00
Gregory Shtrasberg
9fa6c68fa6 [ROCm] Enabling encoder and encoder-decoder on ROCm and AITER unified backends (#35334) 
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
 2026-02-27 21:32:55 +00:00