Commits · 57e9bf18642a391e918400a5afc7c01221635698 · OpenDAS / vllm_cscc

12 Dec, 2025 1 commit
- [Attention] Use sparse prefill kernel for fp8 kv-cache in DeepSeek-v3.2 (#27532) · 3e41992f
  Lucas Wilkinson authored Dec 12, 2025
```
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
```
  3e41992f
11 Dec, 2025 1 commit
- [perf] Use direct copy (broadcast) instead of cat for k_nope/k_pe in MLA prefill (#29710) · fba89069
  Ming Yang authored Dec 11, 2025
```
Signed-off-by: Ming Yang <minos.future@gmail.com>
```
  fba89069
09 Dec, 2025 1 commit
- [DCP][Bugfix][CI] Fix accuracy issue of DCP when using FLASH_ATTN_MLA (#30309) · 67475a6e
  Jaya Yuan authored Dec 09, 2025
```
Signed-off-by: FENP <yuanyongjie.yyj@antgroup.com>
```
  67475a6e
08 Dec, 2025 1 commit
- [Perf] Improve fp8 quant in mla; replace ReduceSum with ReduceScatterSum (#29795) · 1fb632fd
  Lain authored Dec 08, 2025
```
Signed-off-by: Siyuan Fu <siyuanf@nvidia.com>
```
  1fb632fd
05 Dec, 2025 1 commit

[Attention][UX][1/N] Add AttentionConfig and change attention env vars to CLI arguments (#26315) · 66e674cd

Matthew Bonanni authored Dec 05, 2025

Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>

66e674cd

30 Nov, 2025 1 commit
- Fix AttributeError about _use_fi_prefill (#29734) · 82c795d6
  Huamin Li authored Nov 29, 2025
```
Signed-off-by: Huamin Li <3ericli@gmail.com>
```
  82c795d6
28 Nov, 2025 1 commit
- bugfix: correct attn output with base 2 or e (#28840) · 9726e645
  Augusto Yao authored Nov 29, 2025
```
Signed-off-by: augusto.yjh <augusto.yjh@antgroup.com>
```
  9726e645
26 Nov, 2025 1 commit
- [Performance][MLA][ROCm] Remove redundant D2D copy in deepseek (#27457) · d9d342d2
  Pleaplusone authored Nov 26, 2025
```
Signed-off-by: ganyi <ygan@amd.com>
```
  d9d342d2
25 Nov, 2025 2 commits

[ROCm][MLA] enable fp8 MLA decode on ROCm (#28032) · cb7214d8

gbyu-amd authored Nov 25, 2025


Signed-off-by: guanbao <gyu@amd.com>
Signed-off-by: Guanbao Yu <gyu@amd.com>
Signed-off-by: gbyu-amd <Guanbao.Yu@amd.com>
Co-authored-by: guanbao <gyu@amd.com>

cb7214d8

[Perf][Deepseek] optimize gather_and_maybe_dequant_cache kernel's perf for... · 77e10c9c

Pleaplusone authored Nov 25, 2025


[Perf][Deepseek] optimize gather_and_maybe_dequant_cache kernel's perf for extremely long sequence (#28029)
Signed-off-by: ganyi <ygan@amd.com>

77e10c9c

22 Nov, 2025 1 commit
- [Attention] Refactor FA `block_size` limitations to hybrid models only (#29084) · 066209a0
  Nicolò Lucchesi authored Nov 22, 2025
```
Signed-off-by: NickLucche <nlucches@redhat.com>
```
  066209a0
20 Nov, 2025 3 commits
- [KVConnector][Core] Support cross-layer KV blocks (#27743) · 64746471
  Or Ozeri authored Nov 20, 2025
```
Signed-off-by: Or Ozeri <oro@il.ibm.com>
```
  64746471
- [ROCm] Add AMD GPU support on Deepseek v3.2 and SparseMLA (#26670) · 06c20c99
  Pleaplusone authored Nov 20, 2025
```
Signed-off-by: ganyi <ygan@amd.com>
```
  06c20c99
- [AMD] Use Decoupled Kernel Block Size to Support AITER MLA block_size=1 (#27715) · 3fb0d909
  Qiang Zhang authored Nov 20, 2025
```
Signed-off-by: chiangzhang <chiangzhang@tencent.com>
```
  3fb0d909
19 Nov, 2025 3 commits

[Feature] Prefill Context Parallel (PCP) basic support (#28718) · 2fd893b4

Qiu authored Nov 20, 2025


Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
Signed-off-by: FENP <yuanyongjie.yyj@antgroup.com>
Signed-off-by: LookAround <lixushi@huawei.com>
Signed-off-by: Jingchun Gao <gaojingchun1@huawei.com>
Signed-off-by: zhenwenqi2024 <zhenwenqi_2022@qq.com>
Co-authored-by: FENP <yuanyongjie.yyj@antgroup.com>
Co-authored-by: LookAround <lixushi@huawei.com>
Co-authored-by: Jingchun Gao <gaojingchun1@huawei.com>
Co-authored-by: zhenwenqi2024 <zhenwenqi_2022@qq.com>
Co-authored-by: Jingchun Gao <63247409+gjc0824@users.noreply.github.com>

2fd893b4

Upstreaming aiter triton attention backend as a new backend (#28701) · ac10fd3c
Aleksandr Malyshev authored Nov 19, 2025
```
Signed-off-by: Aleksandr Malyshev <maleksan@amd.com>
Co-authored-by: Aleksandr Malyshev <maleksan@amd.com>
```
ac10fd3c
[BugFix] Fix async-scheduling + FlashAttn MLA (#28990) · 48fc8b1e
Lucas Wilkinson authored Nov 19, 2025
```
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
```
48fc8b1e

13 Nov, 2025 2 commits
- [bugfix] correct local_chunk_len for DCP in reorg_kvcache with long context (#28526) · 968060c1
  Qiu authored Nov 14, 2025
```
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
```
  968060c1
- [CI Failure] Fix backend selection for encoder-only models (#28534) · 07a606aa
  Huamin Li authored Nov 13, 2025
```
Signed-off-by: Huamin Li <3ericli@gmail.com>
```
  07a606aa
12 Nov, 2025 2 commits

[Perf] Refactor cudagraph_support to enable full CUDA graphs for spec decoding... · 30441957

Benjamin Chislett authored Nov 12, 2025


[Perf] Refactor cudagraph_support to enable full CUDA graphs for spec decoding with FlashInfer (#28479)
Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>

30441957

`VLLM_USE_TRITON_FLASH_ATTN` V0 variable deprecation (#27611) · 9f0247cf

Andreas Karatzas authored Nov 11, 2025


Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>

9f0247cf

11 Nov, 2025 3 commits

[Feature] Allow configuring FlashInfer workspace size (#28269) · 412e153d

Max Hu authored Nov 11, 2025


Signed-off-by: Max Hu <hyoung2991@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

412e153d

[Misc] fix typo in DCP comment (#28389) · 8c32c6e4
Jie Luo authored Nov 12, 2025
```
Signed-off-by: Livinfly <luojie3m@gmail.com>
```
8c32c6e4

[Attention] Refactor CUDA attention backend selection logic (#24794) · b30dfa03

Matthew Bonanni authored Nov 11, 2025


Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>

b30dfa03

10 Nov, 2025 2 commits

[RFC][ROCm][AITER] Keep all AITER kernels in `_aiter_ops` class like... · f080a835

vllmellm authored Nov 10, 2025


[RFC][ROCm][AITER] Keep all AITER kernels in `_aiter_ops` class like `_custom_ops` and `_ipex_ops` (#24490)
Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>

f080a835

[V0 deprecation] Remove no longer used `get_metadata_cls` (#28370) · e8697faf
Lucas Wilkinson authored Nov 10, 2025
```
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
```
e8697faf

08 Nov, 2025 2 commits

[DCP] Support dcp kv_cache interleave size > 1 (#26696) · 2108a571

zhangsicheng5 authored Nov 09, 2025


Signed-off-by: zhangsicheng5 <zhangsicheng5@huawei.com>
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
Signed-off-by: Qiu <qiuchunshuo@huawei.com>
Co-authored-by: QiuChunshuo <qiuchunshuo@huawei.com>

2108a571

[Attention] Remove max cudagraph size limit of 992 (#27840) · 608bb144
22quinn authored Nov 07, 2025
```
Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>
```
608bb144

05 Nov, 2025 3 commits
- [ROCm][MLA] Support block-size > 1 for AITER MLA backend (#27224) · 6cae1e53
  Pleaplusone authored Nov 05, 2025
```
Signed-off-by: ganyi <ygan@amd.com>
Co-authored-by: wuhuikx <hattie.wu@amd.com>
```
  6cae1e53
- [bugfix] fix wrong `dcp_local_seq_lens` calc (#27518) · 16b37f31
  Qiu authored Nov 05, 2025
```
Signed-off-by: Qiu <qiuchunshuo@huawei.com>
```
  16b37f31
- [BugFix] Fix DCP Assert (AssertionError: DCP not support reorder_batch_threshold > 1 now.) (#28100) · d43ad5a7
  Lucas Wilkinson authored Nov 05, 2025
```
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
```
  d43ad5a7
04 Nov, 2025 1 commit
- [Bug] Batch invariant: Fix flash attn MLA `RuntimeError: scheduler_metadata... · 7e4be741
  Wentao Ye authored Nov 04, 2025
```
[Bug] Batch invariant: Fix flash attn MLA `RuntimeError: scheduler_metadata must have shape (metadata_size)` (#27884)
```
  7e4be741
03 Nov, 2025 1 commit
- [Bugfix] change FlashMLA reorder_batch_threshold (#27777) · 145c00a4
  Matthew Bonanni authored Nov 03, 2025
```
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
```
  145c00a4
31 Oct, 2025 1 commit
- Add FLASHINFER_MLA to test_mla_backends and add B200 CI run (#27663) · f29aeb5a
  Matthew Bonanni authored Oct 31, 2025
```
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
```
  f29aeb5a
26 Oct, 2025 1 commit

[Chore]:Extract math and argparse utilities to separate modules (#27188) · 71b1c8b6

Yeshwanth N authored Oct 26, 2025


Signed-off-by: Yeshwanth Surya <yeshsurya@gmail.com>
Signed-off-by: Yeshwanth N <yeshsurya@gmail.com>
Signed-off-by: yeshsurya <yeshsurya@gmail.com>

71b1c8b6

24 Oct, 2025 2 commits

[Attention] Add MLA prefill backend: trtllm_ragged_attention_deepseek (#26397) · 0f67d4d9
Ming Yang authored Oct 24, 2025
```
Signed-off-by: Ming Yang <minos.future@gmail.com>
```
0f67d4d9

[MISC] `cudagraph_capture_sizes` related improvements (#26016) · 284cc922

fhl2000 authored Oct 24, 2025

Signed-off-by: fhl <2410591650@qq.com>
Signed-off-by: fhl2000 <63384265+fhl2000@users.noreply.github.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

284cc922

23 Oct, 2025 1 commit
- [Attention] Fix FlashMLA metadata builder arguments for q_len > 1 (#27368) · dbfbf9f3
  Matthew Bonanni authored Oct 23, 2025
```
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
```
  dbfbf9f3
22 Oct, 2025 1 commit
- [BugFix] bugfix for Flash Attention MLA with full cuda graph IMA following pr-25490 (#27128) · 5beacce2
  Daisy-Ma-coder authored Oct 22, 2025
```
Signed-off-by: qqma <qqma@amazon.com>
Co-authored-by: qqma <qqma@amazon.com>
```
  5beacce2
21 Oct, 2025 1 commit

[Bugfix] fixes the decoding metadata of dense mla's fp8 kvcache. (#27144) · 250fb1b8

Tao He authored Oct 22, 2025


Signed-off-by: Tao He <linzhu.ht@alibaba-inc.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>

250fb1b8