Commits · dae68969774e41b93b01cd31171ca033a92b574a · OpenDAS / vllm_cscc

07 Mar, 2025 1 commit

[Perf] Reduce MLA CPU overheads in V1 (#14384) · dae68969

Lucas Wilkinson authored Mar 06, 2025


Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>

dae68969

06 Mar, 2025 23 commits
- [Hardware][TPU]Enable ragged paged attention kernel and resolve recompilation issue (#14310) · 0578e5a4
  Chengji Yao authored Mar 06, 2025
```
Signed-off-by: Chengji Yao <chengjiyao@google.com>
```
  0578e5a4
- [V1][Bugfix] Standardize quantized kv cache rejection for attention backends (#14221) · 6832707e
  Michael Goin authored Mar 06, 2025
```
Signed-off-by: mgoin <mgoin64@gmail.com>
```
  6832707e
- [Bug] Fix Attention when ignored in by quant_method (#14313) · 6b2ef5cd
  Michael Goin authored Mar 06, 2025
```
Signed-off-by: mgoin <mgoin64@gmail.com>
```
  6b2ef5cd
- [Bugfix] Fix use_direct_call condition in FusedMoE layer for (#14382) · 958adce4
  Tyler Michael Smith authored Mar 06, 2025
```
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
```
  958adce4
- [Distributed] Add enable_expert_parallel arg (#14305) · cc2f9b32
  Tyler Michael Smith authored Mar 06, 2025
```
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
```
  cc2f9b32
- [V1] Do not detokenize if sampling param detokenize is False (#14224) · cd579352
  Himanshu Jaju authored Mar 06, 2025
```
Signed-off-by: Himanshu Jaju <hj@mistral.ai>
Signed-off-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
```
  cd579352
- Fix mla prefill context performance (#13897) · 9f1710f1
  Ying Zhong authored Mar 07, 2025
```
Signed-off-by: ZhongYingMatrix <zhongyingmatrix@gmail.com>
```
  9f1710f1
- Add authors to license header. (#14371) · e642ec96
  Thomas Parnell authored Mar 06, 2025
```
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Co-authored-by: Burkhard Ringlein <ngl@zurich.ibm.com>
Co-authored-by: Jan van Lunteren <jvl@zurich.ibm.com>
```
  e642ec96
- Reinstate `best_of` for V0 (#14356) · bf0560bd
  Harry Mellor authored Mar 06, 2025
```
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
```
  bf0560bd
- [RLHF] use worker_extension_cls for compatibility with V0 and V1 (#14185) · 151b08e0
  youkaichao authored Mar 07, 2025
```
Signed-off-by: youkaichao <youkaichao@gmail.com>
```
  151b08e0
- [Core] Don't use cache during multi-modal profiling (#14336) · 82551ad6
  Cyrus Leung authored Mar 07, 2025
  
  82551ad6
- [Bugfix][Core] fix abort_seq_group and memory leak when n>1 (#14326) · caac5c2e
  courage17340 authored Mar 06, 2025
```
Signed-off-by: courage17340 <courage17340@163.com>
```
  caac5c2e
- [Kernel] [V1] Improved performance for V1 Triton (ROCm) backend (#14152) · 6bd1dd9d
  Thomas Parnell authored Mar 06, 2025
  
  6bd1dd9d
- [Frontend][Docs] Transcription API streaming (#13301) · fa82b938
  Nicolò Lucchesi authored Mar 06, 2025
```
Signed-off-by: NickLucche <nlucches@redhat.com>
```
  fa82b938
- [Core] Optimizing cross-attention `QKVParallelLinear` computation (#12325) · 69ff99fd
  Nicolò Lucchesi authored Mar 06, 2025
```
Signed-off-by: NickLucche <nlucches@redhat.com>
Signed-off-by: NickLucche <nick@nlucches-4xa100.c.openshift-330514.internal>
Co-authored-by: NickLucche <nick@nlucches-4xa100.c.openshift-330514.internal>
```
  69ff99fd
- [V1][VLM][Pixtral-HF] Support Pixtral-HF on V1 (#14275) · 5d802522
  lkchen authored Mar 06, 2025
```
Signed-off-by: Linkun Chen <github@lkchen.net>
```
  5d802522
- [Model] Update Paligemma multimodal processing with PromptUpdate (#14015) · 17699280
  kYLe authored Mar 06, 2025
```
Signed-off-by: Kyle Huang <kylhuang@nvidia.com>
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
```
  17699280
- [Hardware] Update the flash attn tag to support Blackwell (#14244) · ed6ea065
  Pavani Majety authored Mar 05, 2025
  
  ed6ea065
- [Bugfix][CI] ALiBi test case in xformers multi_query_kv_attention (#11301) · 5ee10e99
  Nicolò Lucchesi authored Mar 06, 2025
  
  5ee10e99
- [Bugfix][Structured Output] Support outlines engine with reasoning outputs for DeepSeek R1 (#14114) · f5f7f00c
  Ce Gao authored Mar 06, 2025
  
  f5f7f00c
- [misc] Mention `ray list nodes` command to troubleshoot ray issues (#14318) · abcc61e0
  Rui Qiao authored Mar 05, 2025
```
Signed-off-by: Rui Qiao <ruisearch42@gmail.com>
```
  abcc61e0
- [BugFix] MLA + V1, illegal memory access and accuracy issues (#14253) · f6bb18fd
  Lucas Wilkinson authored Mar 05, 2025
```
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
```
  f6bb18fd
- [BugFix] Fix prefix caching V0 MLA (#14255) · 4dacaa4a
  Lucas Wilkinson authored Mar 05, 2025
```
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Co-authored-by: Ying Zhong <zhongyingmatrix@gmail.com>
```
  4dacaa4a
05 Mar, 2025 16 commits
- [Bugfix] Remove num_tokens_across_dp (#14302) · a7ea35aa
  Tyler Michael Smith authored Mar 05, 2025
```
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
```
  a7ea35aa
- [Bugfix] Fix DeepSeek MTP crash when using TP1ModelRunner with CUDA graph due... · 1e3e76b6
  pyc96 authored Mar 05, 2025
```
[Bugfix] Fix DeepSeek MTP crash when using TP1ModelRunner with CUDA graph due to shape mismatch (#14237)
Signed-off-by: pyc96 <pychen96@gmail.com>
```
  1e3e76b6
- [misc] Add FlashMLA as a new option of VLLM_ATTENTION_BACKEND env (#14267) · 1b7624bf
  Serena authored Mar 06, 2025
  
  1b7624bf
- [V1][BugFix] Fix for mixed top_k batch (#14301) · ac60dc7f
  Nick Hill authored Mar 05, 2025
```
Signed-off-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Ye Cao <caoye.cao@alibaba-inc.com>
```
  ac60dc7f
- Deprecate `best_of` Sampling Parameter in anticipation for vLLM V1 (#13997) · a4f1ee35
  Vincent authored Mar 05, 2025
```
Signed-off-by: vincent-4 <vincentzhongy+githubvincent4@gmail.com>
Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
```
  a4f1ee35
- [V1][Minor] Remove obsolete FIXME comment (#14304) · a32c8669
  Nick Hill authored Mar 05, 2025
```
Signed-off-by: Nick Hill <nhill@redhat.com>
```
  a32c8669
- [LoRA] Remove linear hack outside transformers backend (#14177) · e17e4488
  Isotr0py authored Mar 05, 2025
```
Signed-off-by: Isotr0py <2037008807@qq.com>
```
  e17e4488
- [V1][Frontend] Add Testing For V1 Runtime Parameters (#14159) · 257e200a
  Robert Shaw authored Mar 05, 2025
```
Signed-off-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
```
  257e200a
- Small update for external_launcher backend docs (#14288) · 47d4a7e0
  Zhe Zhang authored Mar 05, 2025
  
  47d4a7e0
- [Bugfix][V1] Fix allowed_token_ids for v1 Sampler (#14169) · 8d6cd32b
  Lu Fang authored Mar 05, 2025
```
Signed-off-by: Lu Fang <lufang@fb.com>
```
  8d6cd32b
- [Misc][V1] Avoid using `envs.VLLM_USE_V1` in mm processing (#14256) · ec79b67c
  Roger Wang authored Mar 04, 2025
```
Signed-off-by: Roger Wang <ywang@roblox.com>
```
  ec79b67c
- [Frontend] Allow return_tokens_as_token_ids to be passed as a request param (#14066) · 32985bed
  Benjamin Chislett authored Mar 05, 2025
```
Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai>
```
  32985bed
- [platforms] improve rocm debugging info (#14257) · 6eaf9302
  youkaichao authored Mar 05, 2025
  
  6eaf9302
- [V1] EP/TP MoE + DP Attention (#13931) · 72c62eae
  Tyler Michael Smith authored Mar 05, 2025
  
  72c62eae
- [Model] New model support for Phi-4-multimodal-instruct (#14119) · 0a995d54
  Congcong Chen authored Mar 04, 2025
  
  0a995d54
- [V1][Bugfix] Do not reset prefix caching metrics (#14235) · ade3f7d9
  Cody Yu authored Mar 04, 2025
  
  ade3f7d9