xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 04:54:56 +08:00
Code Issues Packages Projects Releases Wiki Activity
10,518 Commits 155 Branches 89 Tags
Commit Graph

10518 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
dongbo910220
b0cde8866e feat(v1): Implement pinned prefix caching with global unpin API 
Core Features:
- Add pin_prefix parameter to SamplingParams for per-request prefix pinning
- Implement pinned prefix caching in V1 engine KVCacheManager
- Add pinned_prefix_cap_ratio (default 0.2) to control memory usage
- Add enable_pinned_prefix global gate for conservative rollouts
- Protect pinned blocks from LRU eviction in BlockPool

Bug Fixes:
- Fix multi-group budget bug with round-robin pinning strategy
- Ensure global cap is never exceeded even with multiple KV cache groups
- Use logical pinned depth (min across groups) for accurate reporting

Management APIs:
- Add HTTP endpoint POST /unpin_all_pinned_prefixes for memory reclamation
- Implement complete call chain: API -> AsyncLLM -> EngineCore -> Scheduler -> KVCacheManager
- Remove per-request unpin to keep API surface minimal

Code Quality:
- Replace manual @field_validator with Field(ge=0, le=1) for cleaner validation
- Add comprehensive test coverage (unit + integration + E2E)
- Add test_multi_group_prefix_pinning_respects_global_cap() for multi-group validation
- Add test_unpin_all_pinned_prefixes_clears_pool() for unpin API validation

Resolves: #23083
Signed-off-by: dongbo910220 <1275604947@qq.com>
 2025-10-17 19:38:21 +08:00
rongfu.leng
5afd3276df
[Feature] Add process_weights_after_loading to AttentionImpl (#26870) 
Signed-off-by: rongfu.leng <rongfu.leng@daocloud.io>
 2025-10-16 08:02:30 -07:00
Tahsin Tunan
43721bc67f
[CI] Replace large models with tiny alternatives in tests (#24057) 
Signed-off-by: Tahsin Tunan <tahsintunan@gmail.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
 2025-10-16 15:51:27 +01:00
Kay Yan
02d709a6f1
[docs] standardize Hugging Face env var to HF_TOKEN (deprecates HUGGING_FACE_HUB_TOKEN) (#27020) 
Signed-off-by: Kay Yan <kay.yan@daocloud.io>
 2025-10-16 15:31:02 +01:00
Mark McLoughlin
4a510ab487
[NIXL] Improve request_finished() debug logs (#25665) 
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
 2025-10-16 15:55:17 +02:00
Matthew Bonanni
314fa8abbf
[Attention] Tune CUTLASS MLA num_splits (#26846) 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
 2025-10-16 06:36:09 -07:00
Cyrus Leung
334535b6fb
[Benchmark] Show E2EL by default for pooling models (#27014) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-16 12:47:09 +00:00
bogdanm
dcbb3f1871
[Bugfix] Correct LayerNorm epsilon parameter in modernbert.py (#27008) 
Signed-off-by: bogdanm <152898065+bogdan01m@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
 2025-10-16 12:27:44 +00:00
Sungjae Lee
00417f4e44
[MISC] fix import violations for re and triton modules (#26654) 
Signed-off-by: Sungjae Lee <33976427+llsj14@users.noreply.github.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
 2025-10-16 03:38:27 -07:00
Lukas Geiger
ed344f4116
Cleanup code after Python 3.10 upgrade (#26520) 
Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com>
 2025-10-16 03:38:23 -07:00
CSWYF3634076
e51928793e
[Model][Bugfix] fix ernie45 vl run failed from shared experts optimization (#26885) 
Signed-off-by: wangyafeng <wangyafeng@baidu.com>
 2025-10-16 03:37:35 -07:00
Cyrus Leung
d2740fafbf
[Chore] Separate out vllm.utils.collections (#26990) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-16 08:35:35 +00:00
Cyrus Leung
17838e50ef
[Benchmark] Use truncation by default for pooling benchmarks (#26992) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-16 16:02:39 +08:00
Zhewen Li
44c8555621
[CI/Build] Fix AMD import failures in CI (#26841) 
Signed-off-by: zhewenli <zhewenli@meta.com>
 2025-10-16 07:28:20 +00:00
Akash kaothalkar
f7d318de2b
[Hardware][CPU][PowerPC]Disable torch.compile() in toptopk sampling (#26987) 
Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com>
Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com>
 2025-10-15 22:36:59 -07:00
Cyrus Leung
76f0d05bc6
[CI/Build] Update expected beam search output for Phi3V (#26978) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-16 05:12:44 +00:00
Bram Wasti
7d8975de84
Deepseek-v3 Batch Invariant on 8xH100 (#26609) 
Signed-off-by: Bram Wasti <bwasti@meta.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
 2025-10-15 22:06:02 -07:00
Vadim Gimpelson
785d8b6410
[PERF] Qwen3-next MTP speedup (change bool mask indexing to index_select / index_copy to reduce d2h) (#26437) 
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
 2025-10-16 12:18:31 +08:00
Cyrus Leung
f6cdc9a02f
[Chore] Rename utils submodules (#26920) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-16 03:58:13 +00:00
Chendi.Xue
509cdc0370
[DOC][XPU]update feature parity with Intel GPU (#26954) 
Signed-off-by: Chendi Xue <Chendi.Xue@intel.com>
Signed-off-by: Chendi Xue <chendi.xue@intel.com>
 2025-10-15 20:07:10 -07:00
Richard Zou
9b6504c307
[BugFix] Work around graph partition x torch.compile cache issue (#26956) 
Signed-off-by: Richard Zou <zou3519@gmail.com>
 2025-10-15 20:06:11 -07:00
Angela Yi
e19b16dde6
[bugfix] Fix SP + PP without specifying compile size (#26955) 
Signed-off-by: angelayi <yiangela7@gmail.com>
 2025-10-15 20:05:33 -07:00
ahao-anyscale
582f2c6be7
[BUG] Allow runai_streamer_sharded in config check (#26958) 
Signed-off-by: ahao-anyscale <ahao@anyscale.com>
 2025-10-15 20:05:14 -07:00
Michael Goin
f8a0acbdbe
[CI] Enable Blackwell Llama4 MoE tests (#26731) 
Signed-off-by: mgoin <mgoin64@gmail.com>
 2025-10-15 21:02:57 -06:00
kliuae
1317034379
[ROCm][FEAT] Fuse DeepSeek shared experts into AITER fused_moe ops (#24097) 
Signed-off-by: chenjun <junchen2@amd.com>
Signed-off-by: kliuae <kuanfu.liu@embeddedllm.com>
Co-authored-by: valarLip <103567126+valarLip@users.noreply.github.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
 2025-10-16 10:41:34 +08:00
InChang Jeong
0ecc553ee6
[Bugfix] reasoning_parser parameter handling in run_batch.py (#26225) 
Signed-off-by: inc-jeong <inc.jeong@navercorp.com>
Signed-off-by: InChang Jeong <inc.jeong@navercorp.com>
Co-authored-by: USER <user@AL02367916.local>
 2025-10-16 10:24:05 +08:00
felixzhu555
f96bc3649c
[Qwen3-Next] Add tuned MoE config for Qwen3-Next FP8 on H100 tp2 (#26887) 
Signed-off-by: Felix Zhu <felixzhu555@gmail.com>
 2025-10-15 18:55:05 -07:00
Alexei-V-Ivanov-AMD
938c43ea7f
[ci] Adjusting AMD test composition 2025-10-14 (#26852) 
Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com>
 2025-10-15 23:52:13 +00:00
Adrian Abeyta
0a9ef0cfce
Move query quantization to attention layer for Flashinfer & Triton. (#26534) 
Signed-off-by: adabeyta <aabeyta@redhat.com>
Signed-off-by: Adrian Abeyta <aabeyta@redhat.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
 2025-10-15 19:01:38 -04:00
Wentao Ye
e5b438a247
[Bug] Temporally Disable VLLM_ALLREDUCE_USE_SYMM_MEM by Default (#26925) 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
 2025-10-15 16:18:50 -04:00
XiaobingZhang
0b99f5d302
support flashinfer_fp4 moe for 5090 gpu (#26669) 
Signed-off-by: XiaobingSuper <xiaobingzhangupc@gmail.com>
Signed-off-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
 2025-10-15 15:06:47 -04:00
Benji Beck
1f491aa0c8
Vectorize RMS norm variance using vectorize_read_with_alignment (#26234) 
Signed-off-by: Benji Beck <benjibeck@meta.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
 2025-10-15 11:54:41 -07:00
Kaixi Hou
de92d916fe
[NVIDIA] Add support for cudnn fp4 gemm via flashinfer (#26107) 
Signed-off-by: kaixih <kaixih@nvidia.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
 2025-10-15 13:53:00 -04:00
Woosuk Kwon
a1063628a4
[Chore] Clean up CODEOWNERS (#26923) 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
 2025-10-15 10:52:54 -07:00
XiaobingZhang
d796375258
[ModelOpt] Remove NVFP4 MoE K%16==0 constraint (#26891) 
Signed-off-by: XiaobingSuper <xiaobingzhangupc@gmail.com>
 2025-10-15 13:06:17 -04:00
Sam/Samuel
14f8456344
[Feature]: Use pydantic validation in observability.py config (#26637) 
Signed-off-by: Samuel Wu <cernunnos1710@gmail.com>
Signed-off-by: Sam/Samuel <57896620+cern1710@users.noreply.github.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
 2025-10-15 16:44:03 +00:00
Pradeep Dasigi
4794c2bd92
Olmo 3 tool parser and tests (#26143) 
Signed-off-by: Pradeep Dasigi <pradeepd@allenai.org>
 2025-10-15 16:36:12 +00:00
Harry Mellor
d3cbaa08dc
Lower sevarity of log when model info cache misses due to exception (#26917) 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
 2025-10-15 09:01:09 -07:00
Cyrus Leung
828523ad8e
[Chore] Separate out vllm.utils.async_utils (#26913) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-15 15:33:00 +00:00
Cyrus Leung
136a17fe6e
[Chore] Separate out vllm.utils.func (#26904) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-15 13:03:58 +00:00
Boyuan Feng
f57438338d
[BugFix] Patch inductor memory plan logic (#26878) 
Signed-off-by: Boyuan Feng <boyuan@meta.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
 2025-10-15 12:51:45 +00:00
Max Wittig
5d598680e3
chore: remove unused marker (#26890) 
Signed-off-by: Max Wittig <max.wittig@siemens.com>
 2025-10-15 05:40:33 -07:00
wangxiyuan
8f4b313c37
[Misc] rename torch_dtype to dtype (#26695) 
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
 2025-10-15 12:11:48 +00:00
Cyrus Leung
f93e348010
[Misc] Remove isort and yapf ignores (#26888) 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
 2025-10-15 12:09:03 +00:00
wang.yuqi
f54f85129e
[Model][2/N] Improve all pooling task | Support multi-vector retrieval (#25370) 
Signed-off-by: wang.yuqi <noooop@126.com>
 2025-10-15 11:14:41 +00:00
li2haipeng
d4d1a6024f
[Lora]Load tuned multi-lora kernel configs from json files (#26319) 
Signed-off-by: li2haipeng <44383182+li2haipeng@users.noreply.github.com>
Signed-off-by: Haipeng Li <li2haipeng@gmail.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
 2025-10-15 09:45:14 +00:00
wangxiyuan
db1764e4e0
[Platform] allow platform to init dp group (#22243) 
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
 2025-10-15 02:32:17 -07:00
Jialin Ouyang
7f83b4ee8e
[Easy] Get rid of unnecessary paraenthesis in kv_cache_manager (#26842) 
Signed-off-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>
 2025-10-15 09:17:43 +00:00
ant-yy
5c3bae1a6a
[Fix] Remove divisibility requirement between num_kv_heads and tp_size in bailing_moe (#26876) 
Signed-off-by: vito.yy <vito.yy@antgroup.com>
 2025-10-15 16:44:04 +08:00
Xudong Ma
5210dc3940
[Misc] Update TritonLanguagePlaceholder to have attributes that are used by Flash Linear Attention ops. (#26853) 
Co-authored-by: Xudong Ma <mxd@meta.com>
 2025-10-15 08:37:49 +00:00