Isotr0py
2db9044ab6
[Bugfix] Fix auto dtype casting for BatchFeature ( #19316 )
...
Signed-off-by: Isotr0py <2037008807@qq.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
2025-06-14 15:13:08 +00:00
Nick Hill
d5bdf899e4
[BugFix] Work-around incremental detokenization edge case error ( #19449 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-06-12 06:43:20 +00:00
Adolfo Victoria
ca27f0f9c1
[Bugfix][Core] Update cancellation logic in generate() to handle Generator exits ( #19225 )
...
Co-authored-by: Adolfo Victoria <adovi@meta.com>
2025-06-06 20:17:54 +00:00
jmswen
7353492a47
[Core] Raise when non-multi-instance DP clients target a DP rank ( #19227 )
...
Signed-off-by: Jon Swenson <jmswen@gmail.com>
2025-06-06 19:03:01 +08:00
jmswen
c8dcc15921
Allow AsyncLLMEngine.generate to target a specific DP rank ( #19102 )
...
Signed-off-by: Jon Swenson <jmswen@gmail.com>
2025-06-04 08:26:47 -07:00
Yan Ru Pei
b712be98c7
feat: add data parallel rank to KVEventBatch ( #18925 )
2025-06-03 17:14:20 -07:00
Simon Mo
02f0c7b220
[Misc] Add SPDX-FileCopyrightText ( #19100 )
...
Signed-off-by: simon-mo <simon.mo@hey.com>
2025-06-03 11:20:17 -07:00
Nick Hill
2dbe8c0774
[Perf] API-server scaleout with many-to-many server-engine comms ( #17546 )
2025-05-30 08:17:00 -07:00
Nick Hill
d1d61f3351
[BugFix] Make DP work with connector-delayed new requests ( #18559 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Will Eaton <weaton@redhat.com>
2025-05-29 18:04:18 +00:00
Mark McLoughlin
06a0338015
[V1][Metrics] Add API for accessing in-memory Prometheus metrics ( #17010 )
...
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
2025-05-27 09:37:06 +00:00
David Xia
de71fec81b
[CI] don't skip fixed test_kv_cache_events() ( #18183 )
...
Signed-off-by: David Xia <david@davidxia.com>
2025-05-14 23:17:16 -07:00
Russell Bryant
78aa341d12
[CI] Fix race condition in test_kv_cache_events test ( #18169 )
...
Signed-off-by: Russell Bryant <rbryant@redhat.com>
2025-05-14 16:27:48 -07:00
Robert Shaw
856865008e
[CI] Disable Failing Tests ( #18165 )
2025-05-14 13:49:56 -07:00
Nick Hill
55aa7af994
[V1] DP scale-out (2/N): Decouple engine process management and comms ( #15977 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-05-13 10:48:21 -07:00
Nick Hill
5ea5c514da
[BugFix] Increase timeout for startup failure test ( #17642 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-05-05 20:53:19 +00:00
Michael Goin
aa4502e7f3
[CI][Bugfix] Fix failing V1 Test due to missing 'cache_salt' arg ( #17500 )
...
Signed-off-by: mgoin <mgoin64@gmail.com>
2025-04-30 21:03:30 -07:00
Alec
0be6d05b5e
[V1][Metrics] add support for kv event publishing ( #16750 )
...
Signed-off-by: alec-flowers <aflowers@nvidia.com>
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Co-authored-by: Mark McLoughlin <markmc@redhat.com>
2025-04-30 07:44:45 -07:00
Marko Rosenmueller
77073c77bc
[Core] Prevent side-channel attacks via cache salting ( #17045 )
...
Signed-off-by: Marko Rosenmueller <5467316+dr75@users.noreply.github.com>
2025-04-30 20:27:21 +08:00
Nick Hill
df6f3ce883
[Core] Remove prompt string from engine core data structures ( #17214 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-04-25 23:41:05 -07:00
Zijing Liu
53e8cf53a4
[V1][Metrics] Allow V1 AsyncLLM to use custom logger ( #14661 )
...
Signed-off-by: Zijing Liu <liuzijing2014@gmail.com>
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Mark McLoughlin <markmc@redhat.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
2025-04-25 22:05:40 -07:00
Rui Qiao
c0dfd97519
[V1][PP] Optimization: continue scheduling prefill chunks ( #17080 )
...
Signed-off-by: Rui Qiao <ruisearch42@gmail.com>
2025-04-24 05:27:08 -07:00
Harry Mellor
0a05ed57e6
Simplify TokenizerGroup ( #16790 )
...
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
2025-04-24 04:43:56 -07:00
Jeffrey Li
0e4254492f
[Bugfix]: fix issue with n>1 sampling on v1 requests overriding each other ( #16863 )
...
Signed-off-by: Jeffrey Li <jeffrey.dot.li@gmail.com>
2025-04-22 11:40:19 +08:00
Nick Hill
7f6d47c1a2
[V1][BugFix] Exit properly if engine core fails during startup ( #16137 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-04-07 15:30:15 -07:00
Cyrus Leung
66d433b94f
[V1] Revert the default max_num_seqs to V0 values for most hardware ( #16158 )
...
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
2025-04-07 13:54:36 -04:00
Nick Hill
15dac210f0
[V1] AsyncLLM data parallel ( #13923 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-03-27 16:14:41 -07:00
Cody Yu
54aa619459
[V1] Refactor num_computed_tokens logic ( #15307 )
...
Signed-off-by: Cody Yu <hao.yu.cody@gmail.com>
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
2025-03-27 04:54:36 +00:00
marko
27df5199d9
Support SHA256 as hash function in prefix caching ( #15297 )
...
Signed-off-by: Marko Rosenmueller <5467316+dr75@users.noreply.github.com>
2025-03-26 11:11:28 -07:00
Nick Hill
9d72daf4ce
[V1][Perf] Simpler request output queues ( #15156 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com>
Co-authored-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com>
2025-03-24 22:44:08 +00:00
Jason
d8e82bc06d
[Bugfix] fix V1 Engine crash while handling requests with duplicate request id ( #15043 )
...
Signed-off-by: Jiahui Sun <jhsun2020@gmail.com>
2025-03-20 10:01:02 -07:00
Murali Andoorveedu
61c7a1b856
[V1] Minor V1 async engine test refactor ( #15075 )
...
Signed-off-by: andoorve <murali.andoorveedu@mail.utoronto.ca>
Co-authored-by: andoorve <murali.andoorveedu@mail.utoronto.ca>
2025-03-19 10:37:17 -07:00
Cyrus Leung
f690372b68
[Core] Update dtype detection and defaults ( #14858 )
...
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
2025-03-19 13:49:33 +08:00
vllmellm
2bb0e1a799
[Bugfix][ROCm] running new process using spawn method for rocm in tests. ( #14810 )
...
Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
2025-03-17 11:33:35 +00:00
Sibi
a73e183e36
[Misc] Replace os environ to monkeypatch in test suite ( #14516 )
...
Signed-off-by: sibi <85477603+t-sibiraj@users.noreply.github.com>
Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: Aaron Pham <contact@aarnphm.xyz>
2025-03-16 20:35:57 -07:00
Robert Shaw
d4d93db2c5
[V1] V1 Enablement Oracle ( #13726 )
...
Signed-off-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Co-authored-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
Co-authored-by: Michael Goin <michael@neuralmagic.com>
2025-03-14 22:02:20 -07:00
afeldman-nm
02fcaa3d0a
[V1] Detokenizer: Respect Stop Tokens + not include_stop_str_in_output ( #14624 )
...
Signed-off-by: Andrew Feldman <afeldman@neuralmagic.com>
2025-03-13 19:07:34 +00:00
Nick Hill
f5d3acd474
[BugFix][V1] Fix parallel sampling finishing/aborts ( #14512 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-03-12 10:29:48 -07:00
afeldman-nm
ef64044079
[V1] Prompt logprobs + APC compatibility; prompt logprobs reqs cannot fill APC ( #13949 )
2025-03-08 01:48:12 +00:00
Nick Hill
8ed5421aaa
[V1] Eagerly remove finished requests from the batch ( #14388 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-03-07 10:56:00 -08:00
Nick Hill
5db6b2c961
[V1][BugFix] Fix remaining sync engine client shutdown errors/hangs ( #13869 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-03-04 15:06:47 +00:00
Harry Mellor
cf069aa8aa
Update deprecated Python 3.8 typing ( #13971 )
2025-03-02 17:34:51 -08:00
afeldman-nm
befc402d34
[V1] V1 engine implements parallel sampling (AsyncLLM and LLMEngine) ( #10980 )
...
Signed-off-by: Andrew Feldman <afeldman@neuralmagic.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
2025-02-24 08:29:41 -08:00
Nick Hill
cbae7af552
[V1][BugFix] Fix engine core client shutdown hangs ( #13298 )
...
Even though ZMQ context.destroy() is meant to close open sockets before terminating the context, it appears to be necessary to do this explicitly or else it can hang in the context.term() method.
Close zmq sockets explicitly before terminating context, make shutdown of client resource more robust, shut down engine core process prior to terminating zmq context.
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-02-23 13:07:43 -08:00
youkaichao
eb24dc4a45
[v1] torchrun compatibility ( #13642 )
...
Signed-off-by: youkaichao <youkaichao@gmail.com>
2025-02-23 22:47:24 +08:00
Nick Hill
caf7ff4456
[V1][Core] Generic mechanism for handling engine utility ( #13060 )
...
Signed-off-by: Nick Hill <nhill@redhat.com>
2025-02-19 17:09:22 +08:00
Murali Andoorveedu
a4d577b379
[V1][Tests] Adding additional testing for multimodal models to V1 ( #13308 )
...
Signed-off-by: andoorve <37849411+andoorve@users.noreply.github.com>
2025-02-18 09:53:14 -08:00
Cody Yu
9206b3d7ec
[V1][PP] Run engine busy loop with batch queue ( #13064 )
2025-02-15 03:59:01 -08:00
Harry Mellor
f2b20fe491
Consolidate Llama model usage in tests ( #13094 )
2025-02-13 22:18:03 -08:00
Mark McLoughlin
75e6e14516
[V1][Metrics] Add several request timing histograms ( #12644 )
...
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
2025-02-11 10:14:00 -05:00
afeldman-nm
0630d4537a
[V1] Logprobs and prompt logprobs support ( #9880 )
...
This PR is adding support for sample logprobs & prompt logprobs to vLLM v1.
New behavior:
- During model execution, model runner computes sample logprobs (if user-provided logprobs setting is not None) and prompt logprobs (if user-provided prompt_logprobs setting is not None). For both sample and prompt logprobs, the engine core returns 3 vectors: token ids, token logprob values, token ranks. Ranks reflect tokens' 1-indexed positions in the vocabulary vector after sorting the vocabulary by log probability in descending order.
- In scheduler.update_from_output(), sample and prompt logprobs are incorporated into the EngineCoreOutput data structure which is transferred to the engine client. If multiprocessing is enabled, then sample and prompt logprobs will be (de)serialized when the EngineCoreOutput data structure is (de)serialized.
- During output processing, the LogprobsProcessor transforms the triplet of token ids, token logprobs values, and token ranks into the OpenAI-compatible List[Dict[token id,Logprob]] format (for sample and prompt logprobs respectively.)
- Each Logprob instance (whether sample- or prompt-) consists of a token's log-probability, rank, and detokenized string representation. Note that logprob detokenization is handled by the LogprobsProcessor not the detokenizer.
Signed-off-by: Andrew Feldman <afeldman@neuralmagic.com>
Signed-off-by: Nick Hill <nhill@redhat.com>
Signed-off-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Co-authored-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
2025-02-07 07:26:20 -08:00
