[Perf] Vectorize static / dynamic INT8 quant kernels #19233
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Hello @yewentao256, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
Summary of Changes
Hello everyone, gemini-code-assist here to provide a summary of this pull request. This PR focuses on performance optimizations for the int8 quantization kernels within vLLM. The primary goal is to improve the throughput and efficiency of models using int8 quantization, addressing issue #18866. The changes involve implementing vectorization techniques in the CUDA kernels responsible for quantization and adding a new benchmark script to measure the performance gains of these kernels.
Highlights
- Kernel Vectorization: The core optimization involves applying vectorization using
vec_n_t(specificallyVEC_SIZE = 16) to the CUDA kernels for static and dynamic int8 quantization (static_scaled_int8_quant_kernel,dynamic_scaled_int8_quant_kernel,dynamic_scaled_int8_azp_quant_kernel). This allows the kernels to process multiple elements simultaneously, improving data throughput on the GPU. - New Benchmark Script: A new benchmark script (
benchmarks/kernels/bench_int8_gemm.py) has been added. This script uses Triton to compare the performance (TFLOP/s) of BF16 GEMM against various int8 GEMM configurations (per-tensor, per-channel, with/without activation quantization time) across different batch sizes. - Performance Improvements: The PR description includes benchmark results showing notable improvements. End-to-end throughput increased from ~35 req/s to ~40 req/s for the tested model. Kernel benchmarks also show increased TFLOP/s for the optimized int8 kernels, particularly at larger batch sizes.
Changelog
- benchmarks/kernels/bench_int8_gemm.py
- Added a new Triton benchmark script to compare BF16 and int8 GEMM kernel performance.
- Includes benchmarking for different int8 quantization schemes (per-tensor, per-channel) and activation quantization timings.
- Configured to run benchmarks for specified models and tensor parallel sizes.
- csrc/quantization/compressed_tensors/int8_quant_kernels.cu
- Included
../vectorization.cuh(line 6). - Implemented vectorization (
VEC_SIZE = 16) instatic_scaled_int8_quant_kernelfor improved data processing (lines 111-142). - Implemented vectorization (
VEC_SIZE = 16) indynamic_scaled_int8_quant_kernelfor the absmax scan and quantization loop (lines 194-251). - Added
MinMaxstruct andMinMaxOpfor efficient min/max reduction in dynamic AZP kernel (lines 254-263). - Implemented vectorization (
VEC_SIZE = 16) indynamic_scaled_int8_azp_quant_kernelfor the min/max scan and quantization loop (lines 269-348).
- Included
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Kernels run fast,
Int8 math, a quick blast,
Vector lanes wide,
Performance takes stride,
Latency's shadow is cast.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request introduces significant performance optimizations to the int8 quantization kernels, backed by comprehensive benchmark results showing notable throughput improvements without accuracy degradation. The core changes involve vectorizing CUDA kernels, which is a standard and effective approach for performance enhancement on GPUs.
The new benchmark script (bench_int8_gemm.py) is a valuable addition for quantifying these improvements across various configurations.
The CUDA kernel modifications in int8_quant_kernels.cu appear solid, incorporating vectorization, pre-calculation of inverse scales, and optimized reduction strategies (like combined Min/Max). These changes are well-aligned with the reported performance gains.
One area for potential clarification/improvement is within the benchmark script concerning the setup for "per-channel" weight quantization scenarios. This is detailed in a specific comment.
Overall, great work on these optimizations!
Summary of Findings
- Benchmark Clarity for Per-Channel Quantization: The benchmark setup for 'per-channel' weight quantization modes (e.g.,
int8-channel-w-token-a) appears to use a single scale factor for quantizing weights, which is then broadcast to a per-channel format for the matrix multiplication. This might not reflect true per-channel quantization where weights are quantized with distinct scales per channel. Clarification on whether this setup is intentional or if it should be adjusted for more realistic per-channel weight quantization benchmarking would be beneficial.
Merge Readiness
The core kernel optimizations in this PR are well-implemented and show significant performance benefits. There is one medium-severity point regarding the setup of 'per-channel' quantization benchmarks that would be good to clarify or address to ensure the benchmarks accurately reflect the intended scenarios. Once this point is resolved, the PR should be in good shape for merging. I am unable to approve the pull request, so please have others review and approve this code before merging.
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
Signed-off-by: yewentao256 <zhyanwentao@126.com>
houseroad
reviewed
Jun 5, 2025
There was a problem hiding this comment.
Looks reasonable to me.
Would leave this to our kernel expert: @chenyang78, @mgoin, @tlrmchlsmth, and @LucasWilkinson to review.
mgoin
reviewed
Jun 6, 2025
There was a problem hiding this comment.
Nice work overall! I'd like to see the PR title and description to have more context on what exactly the changes are to achieve this "optimization", and what hardware it was tested on to give context to the performance numbers.
I also would like to see an expansion to the test cases in vllm/tests/kernels/quantization/test_int8_quant.py to included these vectorization test cases from test_fp8_quant.py
| @@ -107,16 +108,37 @@ template <typename scalar_t, typename scale_type> | |||
| __global__ void static_scaled_int8_quant_kernel( | |||
| scalar_t const* __restrict__ input, int8_t* __restrict__ out, | |||
| scale_type const* scale_ptr, const int hidden_size) { | |||
| int const VEC_SIZE = 16; | |||
There was a problem hiding this comment.
I think it would be best to use constexpr here, in case it matters to the compiler for the #pragma unroll later. At least I did this for fp8 i.e. constexpr size_t VEC_SIZE = 16;
There was a problem hiding this comment.
Fixed, thanks! Made a common function vectorize_with_alignment for it to avoid this mistake
Comment on lines
305
to
309
| // reduce the min and max values across the block in one go | ||
| using BlockReduce = cub::BlockReduce<MinMax, 1024>; | ||
| __shared__ typename BlockReduce::TempStorage reduce_storage; | ||
| MinMax block_min_max = | ||
| BlockReduce(reduce_storage).Reduce(thread_min_max, MinMaxOp()); |
yewentao256
changed the title
yewentao256
changed the title
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
yewentao256
changed the title
@mgoin Thanks for the review, your insight is really valuable! I did one step further and make a common function |
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Comment on lines
+67
to
+71
| # Dynamic per-token quant for A, static per-tensor quant for B | ||
| scale_b = torch.tensor([1.0], device=device, dtype=torch.float32) | ||
| b_int8, scale_b_int8, _ = vllm_scaled_int8_quant(b, scale_b) | ||
| assert scale_b_int8.numel() == 1 | ||
| a_int8, scale_a_int8, _ = vllm_scaled_int8_quant(a) |
There was a problem hiding this comment.
I know this is a bnechmark script but I think this could still be refactored. Perhaps a few functions/objects and a dictionary?
There was a problem hiding this comment.
Great idea! I will have another PR optimizing this, because generally I reuse the code from benchmarks/kernels/bench_fp8_gemm.py and we can update them together
Comment on lines
+94
to
+95
| def run_quant(): | ||
| return vllm_scaled_mm(a_int8, b_int8, scale_a_int8, scale_b_int8, dtype) |
There was a problem hiding this comment.
Why doesn't run_quant include scaled_int8_quant here?
There was a problem hiding this comment.
Because this is for the noquant branch, where we don't measure the time for activations quant (for comparison with the other branch)
Comment on lines
254
to
257
| struct MinMax { | ||
| float min; | ||
| float max; | ||
| }; |
There was a problem hiding this comment.
I like this utility but you could put a lot more of the code in here I think:
- constructor: initialize to
numeric_limits operator&=/operator+=(orvoid reduce): combineMinMaxwith anotherMinMax, or add a value (updates both min and max members).
That way the code that uses it will be much cleaner.
There was a problem hiding this comment.
Great idea! Fixed
| // 2. vectorize the main part | ||
| for (int i = tid; i < num_vec; i += stride) { | ||
| vout_t tmp; | ||
| vec_op(tmp, v_in[i]); |
There was a problem hiding this comment.
Could this just do the for loop here and call the vec_op? Or maybe vec_op can have a default parameter value DefaultVec<VEC_SIZE>{sca_op} that loops the sca_op?
There was a problem hiding this comment.
DefaultVec<VEC_SIZE> looks better for me, fixed
|
|
||
| // 1. prefill the when it is unsafe to vectorize | ||
| for (int i = tid; i < prefix_elems; i += stride) { | ||
| sca_op(out[i], in[i]); |
There was a problem hiding this comment.
Could you make these return vout_t/OutT instead of passing in a parameter?
There was a problem hiding this comment.
Not sure about this, is this functional kernel preferred (return something) or out as a param kernel preferred in vllm?
There was a problem hiding this comment.
I prefer the return, it's more functional, looks more clear, and I think it's a better general c++/CUDA practice, and the compiler is nominally better at optimizing it (in general - here I think everything will be inlined anyway). So I always default to returning.
There was a problem hiding this comment.
I agree with you, and raise another issue for community discussion for a clearer code standard. If the return format is preferred, we can adjust all of the out-pram code in a new pr.
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
mgoin
reviewed
Jun 10, 2025
mgoin
reviewed
Jun 10, 2025
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
mgoin
approved these changes
Jun 10, 2025
mgoin
added
performance
|
@yewentao256 can you please merge with latest main and push to fix some of the failing tests? |
|
Can we keep performance gains off the title of the PR? |
tlrmchlsmth
approved these changes
Jun 12, 2025
|
Nice work |
yewentao256
changed the title
yewentao256
changed the title
Sure, updated, thanks! |
|
Could you force merge into the main? @houseroad |
amogkam
added a commit
to character-tech/vllm
that referenced
this pull request
Jun 16, 2025
* [doc] clarify windows support (vllm-project#19088) Signed-off-by: youkaichao <youkaichao@gmail.com> * [CI/Build] Remove V0 LoRA test (vllm-project#19066) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * Fix underscores in dict keys passed via CLI (vllm-project#19030) Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> * [Bugfix] disable processor cache (vllm-project#19068) Signed-off-by: raushan <raushan@huggingface.co> * [Doc] Improve the Pull Request template with key components (vllm-project#19086) Signed-off-by: Lu Fang <lufang@fb.com> * [Misc] Add missing `_Backend` enums (vllm-project#19081) Signed-off-by: nicklucche <nlucches@redhat.com> * [Misc] fix: add miss best_of param validation (vllm-project#18555) Signed-off-by: googs1025 <googs1025@gmail.com> * [Misc] Add SPDX-FileCopyrightText (vllm-project#19100) Signed-off-by: simon-mo <simon.mo@hey.com> * [Doc] Readme standardization (vllm-project#18695) Co-authored-by: Soren Dreano <soren@numind.ai> * [doc] update docker version (vllm-project#19074) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Kernel] DeepEP dispatch-combine kernel integration (vllm-project#18434) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * [V1] Support cross-layer KV sharing (vllm-project#18212) Signed-off-by: Yong Hoon Shin <yhshin@meta.com> * [Perf] Tune `scaled_fp8_quant` by increasing vectorization (vllm-project#18844) Signed-off-by: mgoin <mgoin64@gmail.com> * Fix interaction between `Optional` and `Annotated` in CLI typing (vllm-project#19093) Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Yikun Jiang <yikun@apache.org> * [v1] Re-init input batch for multiple kv cache groups (vllm-project#18654) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [V1][Spec Decode][Ngram] 1.35x gain -> 1.95x gain on InstructCoder with prompt fix (vllm-project#18971) * [Bugfix] get_num_blocks_to_allocate with null_block (vllm-project#19031) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [Bugfix]: Fix the incompatibility issue with tool_choice 'required' when Thinking is enabled (vllm-project#19075) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Bugfix][P/D] Fix Prefix Cache Bug (vllm-project#18411) Signed-off-by: nicklucche <nlucches@redhat.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> * [Bugfix] Max concurrency estimation and check_enough_kv_cache_memory for models with sliding window layers (vllm-project#19029) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * feat: add data parallel rank to KVEventBatch (vllm-project#18925) * [Misc] Fix path and python alias errors in disagg_prefill exmaples (vllm-project#18919) * [Docs] Add developer doc about CI failures (vllm-project#18782) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Mark McLoughlin <markmc@redhat.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> * [CPU] V1 support for the CPU backend (vllm-project#16441) * [Core] Cast multimodal input in hf processor (vllm-project#18862) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * [KERNEL] Sampler. CUDA kernel for applying repetition penalty (vllm-project#18437) * [Cleanup][v1]:remote guided-decoding-backend for example (vllm-project#19059) Signed-off-by: calvin chen <120380290@qq.com> * [NVIDIA] Add Cutlass MLA backend (vllm-project#17625) * [Bugfix] Fix FA3 full cuda graph correctness (vllm-project#19106) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * Fix vllm-project#19130 (vllm-project#19132) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * [TPU] Skip hanging tests (vllm-project#19115) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * Fix ValueError: Missing value for tag key(s): model_name,engine. (vllm-project#19113) Signed-off-by: Seiji Eicher <seiji@anyscale.com> * [Misc] Add packages for benchmark as extra dependency (vllm-project#19089) Signed-off-by: Isotr0py <2037008807@qq.com> * Improve the output precision of embedding models (vllm-project#19092) * [CI/Build][Bugfix] Ensure compatibility with transformers 4.52 (vllm-project#18678) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Add DeepSeek-R1-0528 function call chat template (vllm-project#18874) Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> * Sm100 blockwise fp8 swap ab (vllm-project#18564) * [Doc] Update V1 Guide for embedding models (vllm-project#19141) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Allow AsyncLLMEngine.generate to target a specific DP rank (vllm-project#19102) Signed-off-by: Jon Swenson <jmswen@gmail.com> * [Bugfix][EP+DP] Fix internode check (vllm-project#19112) Signed-off-by: Tyler Michael Smith <tysmith@redhat.com> * [Perf] Tunings for SM100 FP8 CUTLASS kernel (vllm-project#18778) Signed-off-by: mgoin <mgoin64@gmail.com> * [TPU] Update dynamo dump file name in compilation test (vllm-project#19108) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * [Bugfix] fix v1 cpu worker fails on macOS (vllm-project#19121) * [Kernel] Integrate batched/masked deepgemm kernel (vllm-project#19111) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun <vsundarr@redhat.com> * [Misc] refactor: simplify EngineCoreClient.make_async_mp_client in AsyncLLM (vllm-project#18817) Signed-off-by: googs1025 <googs1025@gmail.com> * [P/D] Heterogeneous TP (vllm-project#18833) Signed-off-by: nicklucche <nlucches@redhat.com> * [doc] small fix (vllm-project#19167) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix][Nixl] Fix full prefix cache hit bug (vllm-project#18632) Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> Signed-off-by: Nick Hill <nhill@redhat.com> Co-authored-by: Nick Hill <nhill@redhat.com> * [Bugfix] Fix port handling in make_zmq_path (vllm-project#19117) * [Torch Nightly]add missing dependency (vllm-project#18770) Signed-off-by: Yang Wang <elainewy@meta.com> * Handle non-serializable objects when dumping benchmark results (vllm-project#19114) * [BugFix][Minor] Fix full cuda graph bug when max_num_seqs < 512 (vllm-project#19171) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [Bugfix]: Fix the incompatibility issue with stream when Thinking is disabled (vllm-project#19135) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Build] Annotate wheel and container path for release workflow (vllm-project#19162) Signed-off-by: simon-mo <simon.mo@hey.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * [Misc] Remove unnecessary fallback to prefill-decode attention (vllm-project#19138) Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com> * [Misc] Do not override NCCL_CUMEM_ENABLE if set explicitly (vllm-project#19105) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Frontend] improve vllm run-batch --help display (vllm-project#19187) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix] properly catch PIL-related errors for vision models when incorrect data urls are provided (vllm-project#19202) Signed-off-by: Guillaume Calmettes <gcalmettes@scaleway.com> * [mistral_common] Add v11 tokenizer (vllm-project#19193) Signed-off-by: Patrick von Platen <patrick.v.platen@gmail.com> * Add H20-3e fused MoE kernel tuning configs for DeepSeek-R1/V3 (vllm-project#19205) * [Hardware][NVIDIA] FP4 MoE kernel optimization (vllm-project#19110) Signed-off-by: Chiyue Wei <chiyuew@nvidia.com> Co-authored-by: Chiyue Wei <chiyuew@nvidia.com> * [MISC][Bugfix] Use less CPU when message queue has been empty for some time (vllm-project#16226) Signed-off-by: Povilas Kanapickas <povilas@radix.lt> * [P/D][NixlConnector] Enable FlashInfer backend (vllm-project#19090) * [Quantization] Skip Fp4 Test for `compressed-tensors` (vllm-project#19217) * [V1] Use FlashInfer by default on Blackwell GPUs (vllm-project#19118) * [Model] NemotronH support (vllm-project#18863) Signed-off-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Co-authored-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> * Fix AOPerModuleConfig name changes (vllm-project#18869) Signed-off-by: Jerry Zhang <jerryzh168@gmail.com> * [Bugfix] Fix EAGLE vocab embedding construction for Llama 70B (vllm-project#19033) Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai> * [v1] Hybrid Memory Allocator (vllm-project#17996) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [TPU] update torch_xla pin (vllm-project#19231) Signed-off-by: Chengji Yao <chengjiyao@google.com> * Support allowed_token_ids in ChatCompletionRequest (vllm-project#19143) Signed-off-by: Xu Song <xusong.vip@gmail.com> * [Chore] update CODEOWNERS (vllm-project#19247) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [v1][P/D] Fix a edge case in kv cache schedule (vllm-project#19182) Co-authored-by: jinghui <jinghui@fb.com> * [TPU] fix kv cache dtype in model runner (vllm-project#19244) Signed-off-by: Chengji Yao <chengjiyao@google.com> * [Quantization] Bump compressed-tensors version; update NVFP4A16 test model (vllm-project#19224) Signed-off-by: Dipika Sikka <dipikasikka1@gmail.com> * [Docs] Improve V1 KVConnector interface documentation (vllm-project#19172) Signed-off-by: Nick Hill <nhill@redhat.com> * Fix CompilationConfig repr (vllm-project#19091) Signed-off-by: rzou <zou3519@gmail.com> * Unit Test for run_dp_sharded_vision_model (vllm-project#19103) Signed-off-by: Siqi Yan <siqi@meta.com> Co-authored-by: Siqi Yan <siqi@meta.com> * [Model] Optimize nemotron_h implementation (vllm-project#19249) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Core] Raise when non-multi-instance DP clients target a DP rank (vllm-project#19227) Signed-off-by: Jon Swenson <jmswen@gmail.com> * improve logits bias (vllm-project#19041) * Fixed ppc build when it runs on non-RHEL based linux distros (vllm-project#18422) Signed-off-by: Nishidha Panpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Signed-off-by: npanpaliya <nishidha.panpaliya@partner.ibm.com> Co-authored-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> * [BugFix] Fix MultiConnector test after HMA changes (vllm-project#19291) Signed-off-by: Nick Hill <nhill@redhat.com> * [Bugfix][Core] Update cancellation logic in `generate()` to handle Generator exits (vllm-project#19225) Co-authored-by: Adolfo Victoria <adovi@meta.com> * [Core] Fix abrupt request abort (vllm-project#18485) Signed-off-by: nicklucche <nlucches@redhat.com> Signed-off-by: Nick Hill <nhill@redhat.com> Co-authored-by: Nick Hill <nhill@redhat.com> * [BugFix] Fix tpu_model_runner block_id concatenation (vllm-project#19228) Signed-off-by: Nick Hill <nhill@redhat.com> * [Misc][Tools][Benchmark] Fix and improve auto tune script (vllm-project#19163) Signed-off-by: Chenyaaang <chenyangli@google.com> * [Build][ROCm] Update Dockerfile.rocm (vllm-project#19296) Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com> * [Easy][Test] Simplify test_function_tool_use with multiple parametrizes (vllm-project#19269) Signed-off-by: Lu Fang <lufang@fb.com> * [Kernel] Integrate CUTLASS MoE kernel with PPLX (vllm-project#18762) Signed-off-by: ElizaWszola <ewszola@redhat.com> Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com> Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com> * [TPU][Test] Add script to run benchmark on TPU for buildkite (vllm-project#19039) Signed-off-by: Qiliang Cui <derrhein@gmail.com> * [CI][PowerPC] Use a more appropriate way to select testcase in tests/models/language/pooling/test_embedding.py (vllm-project#19253) Signed-off-by: Aaruni Aggarwal <aaruniagg@gmail.com> * Add FlexAttention to V1 (vllm-project#16078) Signed-off-by: drisspg <drisspguessous@gmail.com> * [Misc] refactor context extension (vllm-project#19246) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [CI/Build] Improve Llama GGUF test robustness (vllm-project#19287) Signed-off-by: Isotr0py <2037008807@qq.com> * [Nit][Benchmark]Fix example in benchmark_serving_structured_output.py (vllm-project#19311) Signed-off-by: Lifan Shen <lifans@meta.com> * [AMD] Update compatible packaging version (vllm-project#19309) Signed-off-by: pramkuma <Pramendra.Kumar@amd.com> * [BugFix][V1] Fix memory profiling bug (vllm-project#18974) Signed-off-by: luka <luka@neuralmagic.com> * [Bugfix]: Fix TypeError: 'float' object cannot be interpreted as an integer (vllm-project#19283) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Bugfix] Re-enable use_cudagraph in vLLM v1 (vllm-project#19299) Signed-off-by: Richard Zou <zou3519@gmail.com> * [Misc] Change tests/compile to use VLLM_V1 by default (vllm-project#19302) Signed-off-by: rzou <zou3519@gmail.com> * Add H20-3e fused MoE kernel tuning configs for Qwen3-235B-A22B (vllm-project#19315) Signed-off-by: Xu Wenqing <xuwq1993@qq.com> * [Hardware][POWER] Add IBM POWER11 Support to CPU Extension Detection (vllm-project#19082) Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> * [Quantization] Add compressed-tensors NVFP4 support (vllm-project#18312) * [Multi Modal] Add an env var for message queue max chunk bytes (vllm-project#19242) Signed-off-by: yZhen <yZhen@fb.com> Co-authored-by: yZhen <yZhen@fb.com> * [Bugfix] model_max_length should consider max_model_len in tokenizer_config (vllm-project#19201) * [Deprecation] Remove `inputs` arg fallback in Engine classes (vllm-project#18799) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Misc] Add documentation update reminder to PR template (vllm-project#19289) Signed-off-by: Isotr0py <2037008807@qq.com> * [Frontend] Remove unreachable code from llm.py (vllm-project#19288) Signed-off-by: KsuParkhamchuk <k.parkhamchuk@gmail.com> * [Misc] Cleanup compilation tests (vllm-project#19343) Signed-off-by: rzou <zou3519@gmail.com> * [doc] improve ci doc (vllm-project#19307) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Doc] Fix description in the Automatic Prefix Caching design doc (vllm-project#19333) Signed-off-by: cr7258 <chengzw258@163.com> * [CI/Build] Fix LoRA test (vllm-project#19350) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Fix] Allow kernel compilation for CUDA capability 8.7 (vllm-project#19328) Signed-off-by: Conroy Cheers <conroy@corncheese.org> * [CI] Introduce rules for llama auto-label (vllm-project#19323) Signed-off-by: Lu Fang <lufang@fb.com> * [Docs] Fix a bullet list in usage/security.md (vllm-project#19358) Signed-off-by: windsonsea <haifeng.yao@daocloud.io> * [full_graph] Fix query_start_loc padding (vllm-project#19321) Signed-off-by: Yinghai Lu <yinghai@thinkingmachines.ai> * [v1] Add fp32 support to v1 engine through flex attn (vllm-project#19319) Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> * [Misc] Fixes and Optimizations for DeepEP + DeepGEMM combination. (vllm-project#19298) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun <vsundarr@redhat.com> * [Bugfix][Core] Prevent token lengths exceeding `max_model_len` in V0 (vllm-project#19348) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Quantization] Bump compressed-tensors version (vllm-project#19295) Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> * [Frontend] Make TIMEOUT_KEEP_ALIVE configurable through env var (vllm-project#18472) Signed-off-by: liusiqian <liusiqian@tal.com> * [TPU]Fix KV cache sharing tests (vllm-project#19371) * [HOT-FIX] Add `kv_sharing_target_layer_name` argument to cutlass_mla backend (vllm-project#19374) Signed-off-by: Pavani Majety <pmajety@nvidia.com> * [Misc] Fix a config typo in disable_hybrid_kv_cache_manager configuration (vllm-project#19383) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * [V1] Reuse V0's memory_profiling util for gpu worker memory profiling (vllm-project#19312) Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> * [Bugfix] Fix benchmark_moe.py (vllm-project#19016) Signed-off-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> * Use xla flag to improve the quantized model performance (vllm-project#19303) Signed-off-by: Xiongfei Wei <isaacwxf23@gmail.com> * Fix docs/mkdocs/hooks/remove_announcement.py (vllm-project#19382) * [Frontend] Add tqdm_leave_pbar to control progress bar visibility (vllm-project#19357) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Core] Use tuple for kv cache group block ids (vllm-project#19175) Signed-off-by: Nick Hill <nhill@redhat.com> * [Bugfix] Fix modelscope token passed in (vllm-project#19389) Signed-off-by: wangli <wangli858794774@gmail.com> Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Jee Jee Li <pandaleefree@gmail.com> * [Core] Batch multi modal input using pinned memory (vllm-project#19169) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * Add security warning to bug report template (vllm-project#19365) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * [Misc] refactor neuron_multimodal and profiling (vllm-project#19397) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * Add clear documentation around the impact of debugging flag (vllm-project#19369) Signed-off-by: Anna Pendleton <pendleton@google.com> * Automatically bind CPU OMP Threads of a rank to CPU ids of a NUMA node. (vllm-project#17930) Signed-off-by: Tsai, Louie <louie.tsai@intel.com> Co-authored-by: Li, Jiang <bigpyj64@gmail.com> * Revert "[v1] Add fp32 support to v1 engine through flex attn" (vllm-project#19404) * [BugFix][FlashInfer] Fix attention backend interface mismatch with unexpected keyword `use_irope` (vllm-project#19134) Signed-off-by: Yunqiu Guo <guorachel@meta.com> * [BugFix][CPU] Fix CPU CI by ignore collecting test_pixtral (vllm-project#19411) Signed-off-by: jiang.li <jiang1.li@intel.com> * Simplify ep kernels installation (vllm-project#19412) Signed-off-by: youkaichao <youkaichao@gmail.com> * [Misc] Slight improvement of the BNB (vllm-project#19418) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Isotr0py <2037008807@qq.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * [Docs] Note that alternative structured output backends are supported (vllm-project#19426) Signed-off-by: Russell Bryant <rbryant@redhat.com> * [ROCm][V1] Adding ROCm to the list of plaforms using V1 by default (vllm-project#19440) Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com> * [Model] use AutoWeightsLoader for commandr (vllm-project#19399) Signed-off-by: py-andy-c <pychen1017@gmail.com> * Add H20-3e fused MoE kernel tuning configs for Qwen3-235B-A22B-FP8 (vllm-project#19401) Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> * [BugFix] Allow use_cudagraph to work with dynamic VLLM_USE_V1 (vllm-project#19390) Signed-off-by: rzou <zou3519@gmail.com> * [New Model]: Support Qwen3 Embedding & Reranker (vllm-project#19260) * [BugFix] Fix docker build cpu-dev image error (vllm-project#19394) Signed-off-by: niu_he <carlton2tang@gmail.com> * Fix test_max_model_len in tests/entrypoints/llm/test_generate.py (vllm-project#19451) Signed-off-by: Lu Fang <lufang@fb.com> * [CI] Disable failing GGUF model test (vllm-project#19454) Signed-off-by: mgoin <mgoin64@gmail.com> * [Misc] Remove unused `MultiModalHasher.hash_prompt_mm_data` (vllm-project#19422) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * Add fused MOE config for Qwen3 30B A3B on B200 (vllm-project#19455) Signed-off-by: Junhao Li <junhao@ubicloud.com> * Fix Typo in Documentation and Function Name (vllm-project#19442) * [ROCm] Add rules to automatically label ROCm related PRs (vllm-project#19405) Signed-off-by: Lu Fang <lufang@fb.com> * [Kernel] Support deep_gemm for linear methods (vllm-project#19085) Signed-off-by: artetaout <lulala341@gmail.com> * [Doc] Update V1 User Guide for Hardware and Models (vllm-project#19474) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Doc] Fix quantization link titles (vllm-project#19478) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Doc] Support "important" and "announcement" admonitions (vllm-project#19479) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Misc] Reduce warning message introduced in env_override (vllm-project#19476) Signed-off-by: Lu Fang <lufang@fb.com> * Support non-string values in JSON keys from CLI (vllm-project#19471) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Add cache to cuda get_device_capability (vllm-project#19436) Signed-off-by: mgoin <mgoin64@gmail.com> * Fix some typo (vllm-project#19475) Signed-off-by: ximing.wxm <ximing.wxm@antgroup.com> Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com> * Support no privileged mode on CPU for docker and kubernetes deployments (vllm-project#19241) Signed-off-by: Tsai, Louie <louie.tsai@intel.com> * [Bugfix] Update the example code, make it work with the latest lmcache (vllm-project#19453) Signed-off-by: Runzhen Wang <wangrunzhen@gmail.com> * [CI] Update FlashInfer to 0.2.6.post1 (vllm-project#19297) Signed-off-by: mgoin <mgoin64@gmail.com> * [doc] fix "Other AI accelerators" getting started page (vllm-project#19457) Signed-off-by: David Xia <david@davidxia.com> * [Misc] Fix misleading ROCm warning (vllm-project#19486) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Docs] Remove WIP features in V1 guide (vllm-project#19498) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [Kernels] Add activation chunking logic to FusedMoEModularKernel (vllm-project#19168) Signed-off-by: Bill Nell <bnell@redhat.com> * [AMD] [Quantization] Add override flag for attention dtype instead of using kv_cache_dtype trigger (vllm-project#17331) Signed-off-by: Randall Smith <Randall.Smith@amd.com> * [UX] Add Feedback During CUDAGraph Capture (vllm-project#19501) Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> * [CI/Build] Fix torch nightly CI dependencies (vllm-project#19505) Signed-off-by: Richard Zou <zou3519@gmail.com> * [CI] change spell checker from codespell to typos (vllm-project#18711) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [BugFix] Force registration of w8a8_block_fp8_matmul_deepgemm via lazy import (vllm-project#19514) Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * Add Triton Fused MoE kernel config for E=16 on B200 (vllm-project#19518) Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca> * [Frontend] Improve error message in tool_choice validation (vllm-project#19239) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [BugFix] Work-around incremental detokenization edge case error (vllm-project#19449) Signed-off-by: Nick Hill <nhill@redhat.com> * [BugFix] Handle missing sep_token for Qwen3-Reranker in Score API (vllm-project#19522) Signed-off-by: strutive07 <strutive07@gmail.com> * [AMD][Kernel][BugFix] fix test_rocm_compressed_tensors_w8a8 for rocm (vllm-project#19509) Signed-off-by: Randall Smith <Randall.Smith@amd.com> * Fix typo (vllm-project#19525) Signed-off-by: 2niuhe <carlton2tang@gmail.com> * [Security] Prevent new imports of (cloud)pickle (vllm-project#18018) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> * [Bugfix][V1] Allow manual FlashAttention for Blackwell (vllm-project#19492) Signed-off-by: mgoin <mgoin64@gmail.com> * [Bugfix] Respect num-gpu-blocks-override in v1 (vllm-project#19503) Signed-off-by: Jon Swenson <jmswen@gmail.com> * [Quantization] Improve AWQ logic (vllm-project#19431) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Doc] Add V1 column to supported models list (vllm-project#19523) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [V1][NixlConnector] Drop `num_blocks` check (vllm-project#19532) Signed-off-by: NickLucche <nlucches@redhat.com> * [Perf] Vectorize static / dynamic INT8 quant kernels (vllm-project#19233) Signed-off-by: yewentao256 <zhyanwentao@126.com> * Fix TorchAOConfig skip layers (vllm-project#19265) Signed-off-by: mobicham <hicham@mobiuslabs.com> * [torch.compile][ROCm] Fuse quantization onto attention using a torch.compile pass (vllm-project#16756) Signed-off-by: Luka Govedič <lgovedic@redhat.com> Co-authored-by: Sage Moore <sage@neuralmagic.com> * [doc] Make top navigation sticky (vllm-project#19540) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Spec Decode][Benchmark] Generalize spec decode offline benchmark to more methods and datasets (vllm-project#18847) * [Misc] Turn MOE_DP_CHUNK_SIZE into an env var (vllm-project#19506) * [Bugfix] Enforce contiguous input for dynamic_per_token FP8/INT8 quant (vllm-project#19452) Signed-off-by: mgoin <mgoin64@gmail.com> * [Doc] Unify structured outputs examples (vllm-project#18196) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [V1] Resolve failed concurrent structured output requests (vllm-project#19565) Signed-off-by: Russell Bryant <rbryant@redhat.com> * Revert "[Build/CI] Add tracing deps to vllm container image (vllm-project#15224)" (vllm-project#19378) * [BugFix] : Fix Batched DeepGemm Experts (vllm-project#19515) Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * [Bugfix] Fix EAGLE vocab embedding for multimodal target model (vllm-project#19570) Signed-off-by: qizixi <qizixi@meta.com> * [Doc] uses absolute links for structured outputs (vllm-project#19582) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [doc] fix incorrect link (vllm-project#19586) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Misc] Correct broken docs link (vllm-project#19553) Signed-off-by: Zerohertz <ohg3417@gmail.com> * [CPU] Refine default config for the CPU backend (vllm-project#19539) Signed-off-by: jiang1.li <jiang1.li@intel.com> * [Fix] bump mistral common to support magistral (vllm-project#19533) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * [Fix] The zip function in Python 3.9 does not have the strict argument (vllm-project#19549) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * use base version for version comparison (vllm-project#19587) Signed-off-by: Boyuan Feng <boyuan@meta.com> * [torch.compile] reorganize the cache directory to support compiling multiple models (vllm-project#19064) Signed-off-by: youkaichao <youkaichao@gmail.com> * [BugFix] Honor `enable_caching` in connector-delayed kvcache load case (vllm-project#19435) Signed-off-by: Nick Hill <nhill@redhat.com> * [Model] Fix minimax model cache & lm_head precision (vllm-project#19592) Signed-off-by: qingjun <qingjun@minimaxi.com> * [Refactor] Remove unused variables in `moe_permute_unpermute_kernel.inl` (vllm-project#19573) Signed-off-by: yewentao256 <zhyanwentao@126.com> * [doc][mkdocs] fix the duplicate Supported features sections in GPU docs (vllm-project#19606) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [CUDA] Enable full cudagraph for FlashMLA (vllm-project#18581) Signed-off-by: luka <luka@neuralmagic.com> * [Doc] Add troubleshooting section to k8s deployment (vllm-project#19377) Signed-off-by: Anna Pendleton <pendleton@google.com> * [torch.compile] Use custom ops when use_inductor=False (vllm-project#19618) * Adding "AMD: Multi-step Tests" to amdproduction. (vllm-project#19508) Signed-off-by: Yida Wu <yidawu@alumni.cmu.edu> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> * [BugFix] Fix DP Coordinator incorrect debug log message (vllm-project#19624) Signed-off-by: Nick Hill <nhill@redhat.com> * [V1][Metrics] Deprecate metrics with gpu_ prefix for non GPU specific metrics. (vllm-project#18354) Signed-off-by: Saheli Bhattacharjee <saheli@krai.ai> * [Bugfix] Fix the speculative decoding test by setting the target dtype (vllm-project#19633) * [Misc] Modularize CLI Argument Parsing in Benchmark Scripts (vllm-project#19593) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix] Fix auto dtype casting for BatchFeature (vllm-project#19316) Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> * [Hardware][NVIDIA][kernel] Fp4 MOE quant kernel optimization (vllm-project#19500) * Only build CUTLASS MoE kernels on Hopper (vllm-project#19648) * [Bugfix] Don't attempt to use triton if no driver is active (vllm-project#19561) * [Fix] Convert kv_transfer_config from dict to KVTransferConfig (vllm-project#19262) * [Perf] Further tunings for SM100 FP8 CUTLASS kernel (vllm-project#19566) * [Bugfix][2/n] Fix speculative decoding CI - Fix test_ngram_e2e_greedy_correctness (vllm-project#19644) * [Kernel] Raise verbose error and consolidate `num_heads/num_kv_heads` divisibility check (vllm-project#19339) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Benchmark] Refactor benchmark script for fp8 & int8 (vllm-project#19627) Signed-off-by: yewentao256 <zhyanwentao@126.com> * Enable prefix caching with full cuda graphs (vllm-project#19617) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [CI/Build] Fix torch nightly CI dependencies part 2 (vllm-project#19589) * [Misc] Remove duplicate multiproc method setting for CPU platform (vllm-project#19649) Signed-off-by: Isotr0py <2037008807@qq.com> * [MISC] Remove unused variableds in C++ (vllm-project#19609) Signed-off-by: Lu Fang <lufang@fb.com> * [Bugfix][Core] Prefix caching causes incorrect outputs due to outdated ComputedBlocksTracker (vllm-project#18957) Signed-off-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Co-authored-by: 刘全 <quan.liu2@dbappsecurity.com.cn> * [Misc][Frontend] passthrough `bad_words` (vllm-project#19564) Signed-off-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> * [Misc] Fix skipped max-model-len validation when deriving max model length from tokenizer config (vllm-project#19660) Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> * [TPU] support attention head dim smaller than 128 (vllm-project#19620) Signed-off-by: Chengji Yao <chengjiyao@google.com> Co-authored-by: mgoin <mgoin64@gmail.com> * [MISC] typo fix (vllm-project#19672) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [CI] Add mteb testing for rerank models (vllm-project#19344) * [Docs] Move multiproc doc to v1 dir (vllm-project#19651) Signed-off-by: Russell Bryant <rbryant@redhat.com> * [Kernel] GGUF MMVQ kernel for multiple input vectors (vllm-project#18754) Signed-off-by: SzymonOzog <szymon.ozog@gmail.com> * [BugFix] Don't catch BaseException when dumping execute_model errors (vllm-project#19626) Signed-off-by: Nick Hill <nhill@redhat.com> * [DOC] Add reasoning capability to vLLM streamlit code (vllm-project#19557) * [Feature]:Allow for Granite MoE Hybrid models with _only_ shared experts. (vllm-project#19652) Signed-off-by: Shawn Tan <shawntan@ibm.com> * [Bugfix] Fix TP inference for Flex attention backend (vllm-project#19657) Signed-off-by: Isotr0py <2037008807@qq.com> * [MISC] bump huggingface_hub pkg to 0.33.0 (vllm-project#19547) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [Bugfix] fix missing 'finish_reason': null in streaming chat (vllm-project#19662) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Kernels] Use empty for modular MoE workspaces (vllm-project#19667) Signed-off-by: Bill Nell <bnell@redhat.com> * [Model] Add support for MiniMaxM1ForCausalLM (shares architecture with MiniMaxText01ForCausalLM) (vllm-project#19677) Signed-off-by: QscQ <qscqesze@gmail.com> * [V1] Change return type on get_multimodal_embeddings() (vllm-project#19446) Signed-off-by: Russell Bryant <rbryant@redhat.com> --------- Signed-off-by: youkaichao <youkaichao@gmail.com> Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Signed-off-by: raushan <raushan@huggingface.co> Signed-off-by: Lu Fang <lufang@fb.com> Signed-off-by: nicklucche <nlucches@redhat.com> Signed-off-by: googs1025 <googs1025@gmail.com> Signed-off-by: simon-mo <simon.mo@hey.com> Signed-off-by: reidliu41 <reid201711@gmail.com> Signed-off-by: Varun <vsundarr@redhat.com> Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: mgoin <mgoin64@gmail.com> Signed-off-by: Chen Zhang <zhangch99@outlook.com> Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> Signed-off-by: Russell Bryant <rbryant@redhat.com> Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> Signed-off-by: calvin chen <120380290@qq.com> Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> Signed-off-by: Siyuan Liu <lsiyuan@google.com> Signed-off-by: Seiji Eicher <seiji@anyscale.com> Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> Signed-off-by: Jon Swenson <jmswen@gmail.com> Signed-off-by: Tyler Michael Smith <tysmith@redhat.com> Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> Signed-off-by: Nick Hill <nhill@redhat.com> Signed-off-by: Yang Wang <elainewy@meta.com> Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com> Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> Signed-off-by: Guillaume Calmettes <gcalmettes@scaleway.com> Signed-off-by: Patrick von Platen <patrick.v.platen@gmail.com> Signed-off-by: Chiyue Wei <chiyuew@nvidia.com> Signed-off-by: Povilas Kanapickas <povilas@radix.lt> Signed-off-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Signed-off-by: Jerry Zhang <jerryzh168@gmail.com> Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai> Signed-off-by: Chengji Yao <chengjiyao@google.com> Signed-off-by: Xu Song <xusong.vip@gmail.com> Signed-off-by: Aaron Pham <contact@aarnphm.xyz> Signed-off-by: Dipika Sikka <dipikasikka1@gmail.com> Signed-off-by: rzou <zou3519@gmail.com> Signed-off-by: Siqi Yan <siqi@meta.com> Signed-off-by: Nishidha Panpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Signed-off-by: npanpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Chenyaaang <chenyangli@google.com> Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com> Signed-off-by: ElizaWszola <ewszola@redhat.com> Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com> Signed-off-by: Qiliang Cui <derrhein@gmail.com> Signed-off-by: Aaruni Aggarwal <aaruniagg@gmail.com> Signed-off-by: drisspg <drisspguessous@gmail.com> Signed-off-by: Lifan Shen <lifans@meta.com> Signed-off-by: pramkuma <Pramendra.Kumar@amd.com> Signed-off-by: luka <luka@neuralmagic.com> Signed-off-by: Richard Zou <zou3519@gmail.com> Signed-off-by: Xu Wenqing <xuwq1993@qq.com> Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Signed-off-by: yZhen <yZhen@fb.com> Signed-off-by: KsuParkhamchuk <k.parkhamchuk@gmail.com> Signed-off-by: cr7258 <chengzw258@163.com> Signed-off-by: Conroy Cheers <conroy@corncheese.org> Signed-off-by: windsonsea <haifeng.yao@daocloud.io> Signed-off-by: Yinghai Lu <yinghai@thinkingmachines.ai> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> Signed-off-by: liusiqian <liusiqian@tal.com> Signed-off-by: Pavani Majety <pmajety@nvidia.com> Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> Signed-off-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> Signed-off-by: Xiongfei Wei <isaacwxf23@gmail.com> Signed-off-by: wangli <wangli858794774@gmail.com> Signed-off-by: Anna Pendleton <pendleton@google.com> Signed-off-by: Tsai, Louie <louie.tsai@intel.com> Signed-off-by: Yunqiu Guo <guorachel@meta.com> Signed-off-by: jiang.li <jiang1.li@intel.com> Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com> Signed-off-by: py-andy-c <pychen1017@gmail.com> Signed-off-by: niu_he <carlton2tang@gmail.com> Signed-off-by: Junhao Li <junhao@ubicloud.com> Signed-off-by: artetaout <lulala341@gmail.com> Signed-off-by: ximing.wxm <ximing.wxm@antgroup.com> Signed-off-by: Runzhen Wang <wangrunzhen@gmail.com> Signed-off-by: David Xia <david@davidxia.com> Signed-off-by: Bill Nell <bnell@redhat.com> Signed-off-by: Randall Smith <Randall.Smith@amd.com> Signed-off-by: Andy Xie <andy.xning@gmail.com> Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca> Signed-off-by: strutive07 <strutive07@gmail.com> Signed-off-by: 2niuhe <carlton2tang@gmail.com> Signed-off-by: NickLucche <nlucches@redhat.com> Signed-off-by: yewentao256 <zhyanwentao@126.com> Signed-off-by: mobicham <hicham@mobiuslabs.com> Signed-off-by: Luka Govedič <lgovedic@redhat.com> Signed-off-by: qizixi <qizixi@meta.com> Signed-off-by: Zerohertz <ohg3417@gmail.com> Signed-off-by: jiang1.li <jiang1.li@intel.com> Signed-off-by: Boyuan Feng <boyuan@meta.com> Signed-off-by: qingjun <qingjun@minimaxi.com> Signed-off-by: Yida Wu <yidawu@alumni.cmu.edu> Signed-off-by: Saheli Bhattacharjee <saheli@krai.ai> Signed-off-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Signed-off-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Signed-off-by: SzymonOzog <szymon.ozog@gmail.com> Signed-off-by: Shawn Tan <shawntan@ibm.com> Signed-off-by: QscQ <qscqesze@gmail.com> Co-authored-by: youkaichao <youkaichao@gmail.com> Co-authored-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Raushan Turganbay <raushan.turganbay@alumni.nu.edu.kz> Co-authored-by: Lu Fang <30275821+houseroad@users.noreply.github.com> Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com> Co-authored-by: CYJiang <86391540+googs1025@users.noreply.github.com> Co-authored-by: Simon Mo <simon.mo@hey.com> Co-authored-by: SorenDreano <71752785+SorenDreano@users.noreply.github.com> Co-authored-by: Soren Dreano <soren@numind.ai> Co-authored-by: Reid <61492567+reidliu41@users.noreply.github.com> Co-authored-by: reidliu41 <reid201711@gmail.com> Co-authored-by: Varun Sundar Rabindranath <varunsundar08@gmail.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Yong Hoon Shin <48474650+sarckk@users.noreply.github.com> Co-authored-by: Michael Goin <mgoin64@gmail.com> Co-authored-by: Yikun Jiang <yikun@apache.org> Co-authored-by: Chen Zhang <zhangch99@outlook.com> Co-authored-by: Ekagra Ranjan <3116519+ekagra-ranjan@users.noreply.github.com> Co-authored-by: Chauncey <chaunceyjiang@gmail.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> Co-authored-by: Yan Ru Pei <yanrpei@gmail.com> Co-authored-by: Jiaxin Shan <seedjeffwan@gmail.com> Co-authored-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Mark McLoughlin <markmc@redhat.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> Co-authored-by: Li, Jiang <jiang1.li@intel.com> Co-authored-by: Lukas Geiger <lukas.geiger94@gmail.com> Co-authored-by: Vadim Gimpelson <156319763+vadiklyutiy@users.noreply.github.com> Co-authored-by: Calvin Chen <45745657+calvin0327@users.noreply.github.com> Co-authored-by: Kaixi Hou <kaixih@nvidia.com> Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Co-authored-by: 汪志鹏 <wangzhipeng628@gmail.com> Co-authored-by: Siyuan Liu <lsiyuan@google.com> Co-authored-by: Seiji Eicher <58963096+eicherseiji@users.noreply.github.com> Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn> Co-authored-by: wang.yuqi <noooop@126.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk> Co-authored-by: Xu Wenqing <121550081+Xu-Wenqing@users.noreply.github.com> Co-authored-by: Lain <fusiyuan2000@hotmail.com> Co-authored-by: jmswen <jmswen@users.noreply.github.com> Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com> Co-authored-by: Kebe <mail@kebe7jun.com> Co-authored-by: Nick Hill <nhill@redhat.com> Co-authored-by: Yang Wang <elainewy@meta.com> Co-authored-by: Huy Do <huydhn@gmail.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com> Co-authored-by: 22quinn <33176974+22quinn@users.noreply.github.com> Co-authored-by: Guillaume Calmettes <gcalmettes@scaleway.com> Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> Co-authored-by: Chiyue Wei <92623189+dubcyfor3@users.noreply.github.com> Co-authored-by: Chiyue Wei <chiyuew@nvidia.com> Co-authored-by: Povilas Kanapickas <povilas@radix.lt> Co-authored-by: Dipika Sikka <dipikasikka1@gmail.com> Co-authored-by: Luis Vega <vegaluisjose@users.noreply.github.com> Co-authored-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Co-authored-by: Jerry Zhang <jerryzh168@gmail.com> Co-authored-by: Benjamin Chislett <benjamin.chislett@centml.ai> Co-authored-by: Chengji Yao <chengjiyao@google.com> Co-authored-by: Xu Song <xusong.vip@gmail.com> Co-authored-by: Aaron Pham <contact@aarnphm.xyz> Co-authored-by: Jinghui Zhang <jinghuizhang0804@gmail.com> Co-authored-by: jinghui <jinghui@fb.com> Co-authored-by: Richard Zou <zou3519@users.noreply.github.com> Co-authored-by: Siqi Yan <ysq0807@hotmail.com> Co-authored-by: Siqi Yan <siqi@meta.com> Co-authored-by: Yu Guo <82124926+yuguo68@users.noreply.github.com> Co-authored-by: Nishidha <nishidha.panpaliya@partner.ibm.com> Co-authored-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Co-authored-by: Adolfo Victoria <adolfokarim@gmail.com> Co-authored-by: Adolfo Victoria <adovi@meta.com> Co-authored-by: Chenyaaang <42742451+Chenyaaang@users.noreply.github.com> Co-authored-by: Alexei-V-Ivanov-AMD <156011006+Alexei-V-Ivanov-AMD@users.noreply.github.com> Co-authored-by: ElizaWszola <ewszola@redhat.com> Co-authored-by: QiliangCui <derrhein@gmail.com> Co-authored-by: Aaruni Aggarwal <47731267+AaruniAggarwal@users.noreply.github.com> Co-authored-by: Driss Guessous <32754868+drisspg@users.noreply.github.com> Co-authored-by: Lifans <draftbks@gmail.com> Co-authored-by: pramenku <7664080+pramenku@users.noreply.github.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Co-authored-by: Akash kaothalkar <61960177+Akashcodes732@users.noreply.github.com> Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Co-authored-by: jennyyyyzhen <47012288+jennyyyyzhen@users.noreply.github.com> Co-authored-by: yZhen <yZhen@fb.com> Co-authored-by: Kseniya Parkhamchuk <43078183+KsuParkhamchuk@users.noreply.github.com> Co-authored-by: Se7en <chengzw258@163.com> Co-authored-by: Conroy Cheers <conroy@corncheese.org> Co-authored-by: Michael Yao <haifeng.yao@daocloud.io> Co-authored-by: Yinghai Lu <yinghai@thinkingmachines.ai> Co-authored-by: Kyle Sayers <kylesayrs@gmail.com> Co-authored-by: liusiqian-tal <141730978+liusiqian-tal@users.noreply.github.com> Co-authored-by: Pavani Majety <pmajety@nvidia.com> Co-authored-by: Ye (Charlotte) Qi <yeq@meta.com> Co-authored-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> Co-authored-by: XiongfeiWei <isaacwxf23@gmail.com> Co-authored-by: Li Wang <wangli858794774@gmail.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: Anna Pendleton <pendleton@google.com> Co-authored-by: Louie Tsai <louie.tsai@intel.com> Co-authored-by: Li, Jiang <bigpyj64@gmail.com> Co-authored-by: Rachel Guo <35738743+YUNQIUGUO@users.noreply.github.com> Co-authored-by: Isotr0py <2037008807@qq.com> Co-authored-by: Gregory Shtrasberg <156009573+gshtras@users.noreply.github.com> Co-authored-by: py-andy-c <37168711+py-andy-c@users.noreply.github.com> Co-authored-by: niu_he <carlton2tang@gmail.com> Co-authored-by: Junhao Li <junhao@ubicloud.com> Co-authored-by: leopardracer <136604165+leopardracer@users.noreply.github.com> Co-authored-by: artetaout <128046886+artetaout@users.noreply.github.com> Co-authored-by: Ximingwang-09 <72070413+Ximingwang-09@users.noreply.github.com> Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com> Co-authored-by: runzhen <wangrunzhen@gmail.com> Co-authored-by: David Xia <david@davidxia.com> Co-authored-by: bnellnm <49004751+bnellnm@users.noreply.github.com> Co-authored-by: rasmith <Randall.Smith@amd.com> Co-authored-by: Ning Xie <andy.xning@gmail.com> Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca> Co-authored-by: wonjun Jang <strutive07@gmail.com> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Co-authored-by: mobicham <37179323+mobicham@users.noreply.github.com> Co-authored-by: Sage Moore <sage@neuralmagic.com> Co-authored-by: kourosh hakhamaneshi <31483498+kouroshHakha@users.noreply.github.com> Co-authored-by: qizixi <22851944+zixi-qi@users.noreply.github.com> Co-authored-by: Hyogeun Oh (오효근) <ohg3417@gmail.com> Co-authored-by: Boyuan Feng <fby.1994@gmail.com> Co-authored-by: qscqesze <qingjun@minimaxi.com> Co-authored-by: Concurrensee <yida.wu@amd.com> Co-authored-by: Saheli Bhattacharjee <47847054+sahelib25@users.noreply.github.com> Co-authored-by: jiahanc <173873397+jiahanc@users.noreply.github.com> Co-authored-by: Konrad Zawora <kzawora@habana.ai> Co-authored-by: maobaolong <baoloongmao@tencent.com> Co-authored-by: Ilya Markov <markovilya197@gmail.com> Co-authored-by: quanliu <33453350+quanliu1991@users.noreply.github.com> Co-authored-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Co-authored-by: Francesco Bertolotti <f14.bertolotti@gmail.com> Co-authored-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Szymon Ożóg <58388001+SzymonOzog@users.noreply.github.com> Co-authored-by: Navanit Dubey <98005188+Navanit-git@users.noreply.github.com> Co-authored-by: Shawn Tan <shawntan@ibm.com> Co-authored-by: qscqesze <qscqesze@gmail.com>
amogkam
added a commit
to character-tech/vllm
that referenced
this pull request
Jun 16, 2025
* [Bugfix] disable processor cache (vllm-project#19068) Signed-off-by: raushan <raushan@huggingface.co> * [Doc] Improve the Pull Request template with key components (vllm-project#19086) Signed-off-by: Lu Fang <lufang@fb.com> * [Misc] Add missing `_Backend` enums (vllm-project#19081) Signed-off-by: nicklucche <nlucches@redhat.com> * [Misc] fix: add miss best_of param validation (vllm-project#18555) Signed-off-by: googs1025 <googs1025@gmail.com> * [Misc] Add SPDX-FileCopyrightText (vllm-project#19100) Signed-off-by: simon-mo <simon.mo@hey.com> * [Doc] Readme standardization (vllm-project#18695) Co-authored-by: Soren Dreano <soren@numind.ai> * [doc] update docker version (vllm-project#19074) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Kernel] DeepEP dispatch-combine kernel integration (vllm-project#18434) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * [V1] Support cross-layer KV sharing (vllm-project#18212) Signed-off-by: Yong Hoon Shin <yhshin@meta.com> * [Perf] Tune `scaled_fp8_quant` by increasing vectorization (vllm-project#18844) Signed-off-by: mgoin <mgoin64@gmail.com> * Fix interaction between `Optional` and `Annotated` in CLI typing (vllm-project#19093) Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Yikun Jiang <yikun@apache.org> * [v1] Re-init input batch for multiple kv cache groups (vllm-project#18654) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [V1][Spec Decode][Ngram] 1.35x gain -> 1.95x gain on InstructCoder with prompt fix (vllm-project#18971) * [Bugfix] get_num_blocks_to_allocate with null_block (vllm-project#19031) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [Bugfix]: Fix the incompatibility issue with tool_choice 'required' when Thinking is enabled (vllm-project#19075) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Bugfix][P/D] Fix Prefix Cache Bug (vllm-project#18411) Signed-off-by: nicklucche <nlucches@redhat.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> * [Bugfix] Max concurrency estimation and check_enough_kv_cache_memory for models with sliding window layers (vllm-project#19029) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * feat: add data parallel rank to KVEventBatch (vllm-project#18925) * [Misc] Fix path and python alias errors in disagg_prefill exmaples (vllm-project#18919) * [Docs] Add developer doc about CI failures (vllm-project#18782) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Mark McLoughlin <markmc@redhat.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> * [CPU] V1 support for the CPU backend (vllm-project#16441) * [Core] Cast multimodal input in hf processor (vllm-project#18862) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * [KERNEL] Sampler. CUDA kernel for applying repetition penalty (vllm-project#18437) * [Cleanup][v1]:remote guided-decoding-backend for example (vllm-project#19059) Signed-off-by: calvin chen <120380290@qq.com> * [NVIDIA] Add Cutlass MLA backend (vllm-project#17625) * [Bugfix] Fix FA3 full cuda graph correctness (vllm-project#19106) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * Fix vllm-project#19130 (vllm-project#19132) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * [TPU] Skip hanging tests (vllm-project#19115) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * Fix ValueError: Missing value for tag key(s): model_name,engine. (vllm-project#19113) Signed-off-by: Seiji Eicher <seiji@anyscale.com> * [Misc] Add packages for benchmark as extra dependency (vllm-project#19089) Signed-off-by: Isotr0py <2037008807@qq.com> * Improve the output precision of embedding models (vllm-project#19092) * [CI/Build][Bugfix] Ensure compatibility with transformers 4.52 (vllm-project#18678) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Add DeepSeek-R1-0528 function call chat template (vllm-project#18874) Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> * Sm100 blockwise fp8 swap ab (vllm-project#18564) * [Doc] Update V1 Guide for embedding models (vllm-project#19141) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Allow AsyncLLMEngine.generate to target a specific DP rank (vllm-project#19102) Signed-off-by: Jon Swenson <jmswen@gmail.com> * [Bugfix][EP+DP] Fix internode check (vllm-project#19112) Signed-off-by: Tyler Michael Smith <tysmith@redhat.com> * [Perf] Tunings for SM100 FP8 CUTLASS kernel (vllm-project#18778) Signed-off-by: mgoin <mgoin64@gmail.com> * [TPU] Update dynamo dump file name in compilation test (vllm-project#19108) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * [Bugfix] fix v1 cpu worker fails on macOS (vllm-project#19121) * [Kernel] Integrate batched/masked deepgemm kernel (vllm-project#19111) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun <vsundarr@redhat.com> * [Misc] refactor: simplify EngineCoreClient.make_async_mp_client in AsyncLLM (vllm-project#18817) Signed-off-by: googs1025 <googs1025@gmail.com> * [P/D] Heterogeneous TP (vllm-project#18833) Signed-off-by: nicklucche <nlucches@redhat.com> * [doc] small fix (vllm-project#19167) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix][Nixl] Fix full prefix cache hit bug (vllm-project#18632) Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> Signed-off-by: Nick Hill <nhill@redhat.com> Co-authored-by: Nick Hill <nhill@redhat.com> * [Bugfix] Fix port handling in make_zmq_path (vllm-project#19117) * [Torch Nightly]add missing dependency (vllm-project#18770) Signed-off-by: Yang Wang <elainewy@meta.com> * Handle non-serializable objects when dumping benchmark results (vllm-project#19114) * [BugFix][Minor] Fix full cuda graph bug when max_num_seqs < 512 (vllm-project#19171) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [Bugfix]: Fix the incompatibility issue with stream when Thinking is disabled (vllm-project#19135) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Build] Annotate wheel and container path for release workflow (vllm-project#19162) Signed-off-by: simon-mo <simon.mo@hey.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * [Misc] Remove unnecessary fallback to prefill-decode attention (vllm-project#19138) Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com> * [Misc] Do not override NCCL_CUMEM_ENABLE if set explicitly (vllm-project#19105) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Frontend] improve vllm run-batch --help display (vllm-project#19187) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix] properly catch PIL-related errors for vision models when incorrect data urls are provided (vllm-project#19202) Signed-off-by: Guillaume Calmettes <gcalmettes@scaleway.com> * [mistral_common] Add v11 tokenizer (vllm-project#19193) Signed-off-by: Patrick von Platen <patrick.v.platen@gmail.com> * Add H20-3e fused MoE kernel tuning configs for DeepSeek-R1/V3 (vllm-project#19205) * [Hardware][NVIDIA] FP4 MoE kernel optimization (vllm-project#19110) Signed-off-by: Chiyue Wei <chiyuew@nvidia.com> Co-authored-by: Chiyue Wei <chiyuew@nvidia.com> * [MISC][Bugfix] Use less CPU when message queue has been empty for some time (vllm-project#16226) Signed-off-by: Povilas Kanapickas <povilas@radix.lt> * [P/D][NixlConnector] Enable FlashInfer backend (vllm-project#19090) * [Quantization] Skip Fp4 Test for `compressed-tensors` (vllm-project#19217) * [V1] Use FlashInfer by default on Blackwell GPUs (vllm-project#19118) * [Model] NemotronH support (vllm-project#18863) Signed-off-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Co-authored-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> * Fix AOPerModuleConfig name changes (vllm-project#18869) Signed-off-by: Jerry Zhang <jerryzh168@gmail.com> * [Bugfix] Fix EAGLE vocab embedding construction for Llama 70B (vllm-project#19033) Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai> * [v1] Hybrid Memory Allocator (vllm-project#17996) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [TPU] update torch_xla pin (vllm-project#19231) Signed-off-by: Chengji Yao <chengjiyao@google.com> * Support allowed_token_ids in ChatCompletionRequest (vllm-project#19143) Signed-off-by: Xu Song <xusong.vip@gmail.com> * [Chore] update CODEOWNERS (vllm-project#19247) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [v1][P/D] Fix a edge case in kv cache schedule (vllm-project#19182) Co-authored-by: jinghui <jinghui@fb.com> * [TPU] fix kv cache dtype in model runner (vllm-project#19244) Signed-off-by: Chengji Yao <chengjiyao@google.com> * [Quantization] Bump compressed-tensors version; update NVFP4A16 test model (vllm-project#19224) Signed-off-by: Dipika Sikka <dipikasikka1@gmail.com> * [Docs] Improve V1 KVConnector interface documentation (vllm-project#19172) Signed-off-by: Nick Hill <nhill@redhat.com> * Fix CompilationConfig repr (vllm-project#19091) Signed-off-by: rzou <zou3519@gmail.com> * Unit Test for run_dp_sharded_vision_model (vllm-project#19103) Signed-off-by: Siqi Yan <siqi@meta.com> Co-authored-by: Siqi Yan <siqi@meta.com> * [Model] Optimize nemotron_h implementation (vllm-project#19249) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Core] Raise when non-multi-instance DP clients target a DP rank (vllm-project#19227) Signed-off-by: Jon Swenson <jmswen@gmail.com> * improve logits bias (vllm-project#19041) * Fixed ppc build when it runs on non-RHEL based linux distros (vllm-project#18422) Signed-off-by: Nishidha Panpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Signed-off-by: npanpaliya <nishidha.panpaliya@partner.ibm.com> Co-authored-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> * [BugFix] Fix MultiConnector test after HMA changes (vllm-project#19291) Signed-off-by: Nick Hill <nhill@redhat.com> * [Bugfix][Core] Update cancellation logic in `generate()` to handle Generator exits (vllm-project#19225) Co-authored-by: Adolfo Victoria <adovi@meta.com> * [Core] Fix abrupt request abort (vllm-project#18485) Signed-off-by: nicklucche <nlucches@redhat.com> Signed-off-by: Nick Hill <nhill@redhat.com> Co-authored-by: Nick Hill <nhill@redhat.com> * [BugFix] Fix tpu_model_runner block_id concatenation (vllm-project#19228) Signed-off-by: Nick Hill <nhill@redhat.com> * [Misc][Tools][Benchmark] Fix and improve auto tune script (vllm-project#19163) Signed-off-by: Chenyaaang <chenyangli@google.com> * [Build][ROCm] Update Dockerfile.rocm (vllm-project#19296) Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com> * [Easy][Test] Simplify test_function_tool_use with multiple parametrizes (vllm-project#19269) Signed-off-by: Lu Fang <lufang@fb.com> * [Kernel] Integrate CUTLASS MoE kernel with PPLX (vllm-project#18762) Signed-off-by: ElizaWszola <ewszola@redhat.com> Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com> Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com> * [TPU][Test] Add script to run benchmark on TPU for buildkite (vllm-project#19039) Signed-off-by: Qiliang Cui <derrhein@gmail.com> * [CI][PowerPC] Use a more appropriate way to select testcase in tests/models/language/pooling/test_embedding.py (vllm-project#19253) Signed-off-by: Aaruni Aggarwal <aaruniagg@gmail.com> * Add FlexAttention to V1 (vllm-project#16078) Signed-off-by: drisspg <drisspguessous@gmail.com> * [Misc] refactor context extension (vllm-project#19246) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [CI/Build] Improve Llama GGUF test robustness (vllm-project#19287) Signed-off-by: Isotr0py <2037008807@qq.com> * [Nit][Benchmark]Fix example in benchmark_serving_structured_output.py (vllm-project#19311) Signed-off-by: Lifan Shen <lifans@meta.com> * [AMD] Update compatible packaging version (vllm-project#19309) Signed-off-by: pramkuma <Pramendra.Kumar@amd.com> * [BugFix][V1] Fix memory profiling bug (vllm-project#18974) Signed-off-by: luka <luka@neuralmagic.com> * [Bugfix]: Fix TypeError: 'float' object cannot be interpreted as an integer (vllm-project#19283) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Bugfix] Re-enable use_cudagraph in vLLM v1 (vllm-project#19299) Signed-off-by: Richard Zou <zou3519@gmail.com> * [Misc] Change tests/compile to use VLLM_V1 by default (vllm-project#19302) Signed-off-by: rzou <zou3519@gmail.com> * Add H20-3e fused MoE kernel tuning configs for Qwen3-235B-A22B (vllm-project#19315) Signed-off-by: Xu Wenqing <xuwq1993@qq.com> * [Hardware][POWER] Add IBM POWER11 Support to CPU Extension Detection (vllm-project#19082) Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> * [Quantization] Add compressed-tensors NVFP4 support (vllm-project#18312) * [Multi Modal] Add an env var for message queue max chunk bytes (vllm-project#19242) Signed-off-by: yZhen <yZhen@fb.com> Co-authored-by: yZhen <yZhen@fb.com> * [Bugfix] model_max_length should consider max_model_len in tokenizer_config (vllm-project#19201) * [Deprecation] Remove `inputs` arg fallback in Engine classes (vllm-project#18799) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Misc] Add documentation update reminder to PR template (vllm-project#19289) Signed-off-by: Isotr0py <2037008807@qq.com> * [Frontend] Remove unreachable code from llm.py (vllm-project#19288) Signed-off-by: KsuParkhamchuk <k.parkhamchuk@gmail.com> * [Misc] Cleanup compilation tests (vllm-project#19343) Signed-off-by: rzou <zou3519@gmail.com> * [doc] improve ci doc (vllm-project#19307) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Doc] Fix description in the Automatic Prefix Caching design doc (vllm-project#19333) Signed-off-by: cr7258 <chengzw258@163.com> * [CI/Build] Fix LoRA test (vllm-project#19350) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Fix] Allow kernel compilation for CUDA capability 8.7 (vllm-project#19328) Signed-off-by: Conroy Cheers <conroy@corncheese.org> * [CI] Introduce rules for llama auto-label (vllm-project#19323) Signed-off-by: Lu Fang <lufang@fb.com> * [Docs] Fix a bullet list in usage/security.md (vllm-project#19358) Signed-off-by: windsonsea <haifeng.yao@daocloud.io> * [full_graph] Fix query_start_loc padding (vllm-project#19321) Signed-off-by: Yinghai Lu <yinghai@thinkingmachines.ai> * [v1] Add fp32 support to v1 engine through flex attn (vllm-project#19319) Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> * [Misc] Fixes and Optimizations for DeepEP + DeepGEMM combination. (vllm-project#19298) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun <vsundarr@redhat.com> * [Bugfix][Core] Prevent token lengths exceeding `max_model_len` in V0 (vllm-project#19348) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Quantization] Bump compressed-tensors version (vllm-project#19295) Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> * [Frontend] Make TIMEOUT_KEEP_ALIVE configurable through env var (vllm-project#18472) Signed-off-by: liusiqian <liusiqian@tal.com> * [TPU]Fix KV cache sharing tests (vllm-project#19371) * [HOT-FIX] Add `kv_sharing_target_layer_name` argument to cutlass_mla backend (vllm-project#19374) Signed-off-by: Pavani Majety <pmajety@nvidia.com> * [Misc] Fix a config typo in disable_hybrid_kv_cache_manager configuration (vllm-project#19383) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * [V1] Reuse V0's memory_profiling util for gpu worker memory profiling (vllm-project#19312) Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> * [Bugfix] Fix benchmark_moe.py (vllm-project#19016) Signed-off-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> * Use xla flag to improve the quantized model performance (vllm-project#19303) Signed-off-by: Xiongfei Wei <isaacwxf23@gmail.com> * Fix docs/mkdocs/hooks/remove_announcement.py (vllm-project#19382) * [Frontend] Add tqdm_leave_pbar to control progress bar visibility (vllm-project#19357) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Core] Use tuple for kv cache group block ids (vllm-project#19175) Signed-off-by: Nick Hill <nhill@redhat.com> * [Bugfix] Fix modelscope token passed in (vllm-project#19389) Signed-off-by: wangli <wangli858794774@gmail.com> Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Jee Jee Li <pandaleefree@gmail.com> * [Core] Batch multi modal input using pinned memory (vllm-project#19169) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * Add security warning to bug report template (vllm-project#19365) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * [Misc] refactor neuron_multimodal and profiling (vllm-project#19397) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * Add clear documentation around the impact of debugging flag (vllm-project#19369) Signed-off-by: Anna Pendleton <pendleton@google.com> * Automatically bind CPU OMP Threads of a rank to CPU ids of a NUMA node. (vllm-project#17930) Signed-off-by: Tsai, Louie <louie.tsai@intel.com> Co-authored-by: Li, Jiang <bigpyj64@gmail.com> * Revert "[v1] Add fp32 support to v1 engine through flex attn" (vllm-project#19404) * [BugFix][FlashInfer] Fix attention backend interface mismatch with unexpected keyword `use_irope` (vllm-project#19134) Signed-off-by: Yunqiu Guo <guorachel@meta.com> * [BugFix][CPU] Fix CPU CI by ignore collecting test_pixtral (vllm-project#19411) Signed-off-by: jiang.li <jiang1.li@intel.com> * Simplify ep kernels installation (vllm-project#19412) Signed-off-by: youkaichao <youkaichao@gmail.com> * [Misc] Slight improvement of the BNB (vllm-project#19418) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Isotr0py <2037008807@qq.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * [Docs] Note that alternative structured output backends are supported (vllm-project#19426) Signed-off-by: Russell Bryant <rbryant@redhat.com> * [ROCm][V1] Adding ROCm to the list of plaforms using V1 by default (vllm-project#19440) Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com> * [Model] use AutoWeightsLoader for commandr (vllm-project#19399) Signed-off-by: py-andy-c <pychen1017@gmail.com> * Add H20-3e fused MoE kernel tuning configs for Qwen3-235B-A22B-FP8 (vllm-project#19401) Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> * [BugFix] Allow use_cudagraph to work with dynamic VLLM_USE_V1 (vllm-project#19390) Signed-off-by: rzou <zou3519@gmail.com> * [New Model]: Support Qwen3 Embedding & Reranker (vllm-project#19260) * [BugFix] Fix docker build cpu-dev image error (vllm-project#19394) Signed-off-by: niu_he <carlton2tang@gmail.com> * Fix test_max_model_len in tests/entrypoints/llm/test_generate.py (vllm-project#19451) Signed-off-by: Lu Fang <lufang@fb.com> * [CI] Disable failing GGUF model test (vllm-project#19454) Signed-off-by: mgoin <mgoin64@gmail.com> * [Misc] Remove unused `MultiModalHasher.hash_prompt_mm_data` (vllm-project#19422) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * Add fused MOE config for Qwen3 30B A3B on B200 (vllm-project#19455) Signed-off-by: Junhao Li <junhao@ubicloud.com> * Fix Typo in Documentation and Function Name (vllm-project#19442) * [ROCm] Add rules to automatically label ROCm related PRs (vllm-project#19405) Signed-off-by: Lu Fang <lufang@fb.com> * [Kernel] Support deep_gemm for linear methods (vllm-project#19085) Signed-off-by: artetaout <lulala341@gmail.com> * [Doc] Update V1 User Guide for Hardware and Models (vllm-project#19474) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Doc] Fix quantization link titles (vllm-project#19478) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Doc] Support "important" and "announcement" admonitions (vllm-project#19479) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Misc] Reduce warning message introduced in env_override (vllm-project#19476) Signed-off-by: Lu Fang <lufang@fb.com> * Support non-string values in JSON keys from CLI (vllm-project#19471) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Add cache to cuda get_device_capability (vllm-project#19436) Signed-off-by: mgoin <mgoin64@gmail.com> * Fix some typo (vllm-project#19475) Signed-off-by: ximing.wxm <ximing.wxm@antgroup.com> Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com> * Support no privileged mode on CPU for docker and kubernetes deployments (vllm-project#19241) Signed-off-by: Tsai, Louie <louie.tsai@intel.com> * [Bugfix] Update the example code, make it work with the latest lmcache (vllm-project#19453) Signed-off-by: Runzhen Wang <wangrunzhen@gmail.com> * [CI] Update FlashInfer to 0.2.6.post1 (vllm-project#19297) Signed-off-by: mgoin <mgoin64@gmail.com> * [doc] fix "Other AI accelerators" getting started page (vllm-project#19457) Signed-off-by: David Xia <david@davidxia.com> * [Misc] Fix misleading ROCm warning (vllm-project#19486) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Docs] Remove WIP features in V1 guide (vllm-project#19498) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [Kernels] Add activation chunking logic to FusedMoEModularKernel (vllm-project#19168) Signed-off-by: Bill Nell <bnell@redhat.com> * [AMD] [Quantization] Add override flag for attention dtype instead of using kv_cache_dtype trigger (vllm-project#17331) Signed-off-by: Randall Smith <Randall.Smith@amd.com> * [UX] Add Feedback During CUDAGraph Capture (vllm-project#19501) Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> * [CI/Build] Fix torch nightly CI dependencies (vllm-project#19505) Signed-off-by: Richard Zou <zou3519@gmail.com> * [CI] change spell checker from codespell to typos (vllm-project#18711) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [BugFix] Force registration of w8a8_block_fp8_matmul_deepgemm via lazy import (vllm-project#19514) Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * Add Triton Fused MoE kernel config for E=16 on B200 (vllm-project#19518) Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca> * [Frontend] Improve error message in tool_choice validation (vllm-project#19239) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [BugFix] Work-around incremental detokenization edge case error (vllm-project#19449) Signed-off-by: Nick Hill <nhill@redhat.com> * [BugFix] Handle missing sep_token for Qwen3-Reranker in Score API (vllm-project#19522) Signed-off-by: strutive07 <strutive07@gmail.com> * [AMD][Kernel][BugFix] fix test_rocm_compressed_tensors_w8a8 for rocm (vllm-project#19509) Signed-off-by: Randall Smith <Randall.Smith@amd.com> * Fix typo (vllm-project#19525) Signed-off-by: 2niuhe <carlton2tang@gmail.com> * [Security] Prevent new imports of (cloud)pickle (vllm-project#18018) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> * [Bugfix][V1] Allow manual FlashAttention for Blackwell (vllm-project#19492) Signed-off-by: mgoin <mgoin64@gmail.com> * [Bugfix] Respect num-gpu-blocks-override in v1 (vllm-project#19503) Signed-off-by: Jon Swenson <jmswen@gmail.com> * [Quantization] Improve AWQ logic (vllm-project#19431) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Doc] Add V1 column to supported models list (vllm-project#19523) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [V1][NixlConnector] Drop `num_blocks` check (vllm-project#19532) Signed-off-by: NickLucche <nlucches@redhat.com> * [Perf] Vectorize static / dynamic INT8 quant kernels (vllm-project#19233) Signed-off-by: yewentao256 <zhyanwentao@126.com> * Fix TorchAOConfig skip layers (vllm-project#19265) Signed-off-by: mobicham <hicham@mobiuslabs.com> * [torch.compile][ROCm] Fuse quantization onto attention using a torch.compile pass (vllm-project#16756) Signed-off-by: Luka Govedič <lgovedic@redhat.com> Co-authored-by: Sage Moore <sage@neuralmagic.com> * [doc] Make top navigation sticky (vllm-project#19540) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Spec Decode][Benchmark] Generalize spec decode offline benchmark to more methods and datasets (vllm-project#18847) * [Misc] Turn MOE_DP_CHUNK_SIZE into an env var (vllm-project#19506) * [Bugfix] Enforce contiguous input for dynamic_per_token FP8/INT8 quant (vllm-project#19452) Signed-off-by: mgoin <mgoin64@gmail.com> * [Doc] Unify structured outputs examples (vllm-project#18196) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [V1] Resolve failed concurrent structured output requests (vllm-project#19565) Signed-off-by: Russell Bryant <rbryant@redhat.com> * Revert "[Build/CI] Add tracing deps to vllm container image (vllm-project#15224)" (vllm-project#19378) * [BugFix] : Fix Batched DeepGemm Experts (vllm-project#19515) Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * [Bugfix] Fix EAGLE vocab embedding for multimodal target model (vllm-project#19570) Signed-off-by: qizixi <qizixi@meta.com> * [Doc] uses absolute links for structured outputs (vllm-project#19582) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [doc] fix incorrect link (vllm-project#19586) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Misc] Correct broken docs link (vllm-project#19553) Signed-off-by: Zerohertz <ohg3417@gmail.com> * [CPU] Refine default config for the CPU backend (vllm-project#19539) Signed-off-by: jiang1.li <jiang1.li@intel.com> * [Fix] bump mistral common to support magistral (vllm-project#19533) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * [Fix] The zip function in Python 3.9 does not have the strict argument (vllm-project#19549) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * use base version for version comparison (vllm-project#19587) Signed-off-by: Boyuan Feng <boyuan@meta.com> * [torch.compile] reorganize the cache directory to support compiling multiple models (vllm-project#19064) Signed-off-by: youkaichao <youkaichao@gmail.com> * [BugFix] Honor `enable_caching` in connector-delayed kvcache load case (vllm-project#19435) Signed-off-by: Nick Hill <nhill@redhat.com> * [Model] Fix minimax model cache & lm_head precision (vllm-project#19592) Signed-off-by: qingjun <qingjun@minimaxi.com> * [Refactor] Remove unused variables in `moe_permute_unpermute_kernel.inl` (vllm-project#19573) Signed-off-by: yewentao256 <zhyanwentao@126.com> * [doc][mkdocs] fix the duplicate Supported features sections in GPU docs (vllm-project#19606) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [CUDA] Enable full cudagraph for FlashMLA (vllm-project#18581) Signed-off-by: luka <luka@neuralmagic.com> * [Doc] Add troubleshooting section to k8s deployment (vllm-project#19377) Signed-off-by: Anna Pendleton <pendleton@google.com> * [torch.compile] Use custom ops when use_inductor=False (vllm-project#19618) * Adding "AMD: Multi-step Tests" to amdproduction. (vllm-project#19508) Signed-off-by: Yida Wu <yidawu@alumni.cmu.edu> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> * [BugFix] Fix DP Coordinator incorrect debug log message (vllm-project#19624) Signed-off-by: Nick Hill <nhill@redhat.com> * [V1][Metrics] Deprecate metrics with gpu_ prefix for non GPU specific metrics. (vllm-project#18354) Signed-off-by: Saheli Bhattacharjee <saheli@krai.ai> * [Bugfix] Fix the speculative decoding test by setting the target dtype (vllm-project#19633) * [Misc] Modularize CLI Argument Parsing in Benchmark Scripts (vllm-project#19593) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix] Fix auto dtype casting for BatchFeature (vllm-project#19316) Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> * [Hardware][NVIDIA][kernel] Fp4 MOE quant kernel optimization (vllm-project#19500) * Only build CUTLASS MoE kernels on Hopper (vllm-project#19648) * [Bugfix] Don't attempt to use triton if no driver is active (vllm-project#19561) * [Fix] Convert kv_transfer_config from dict to KVTransferConfig (vllm-project#19262) * [Perf] Further tunings for SM100 FP8 CUTLASS kernel (vllm-project#19566) * [Bugfix][2/n] Fix speculative decoding CI - Fix test_ngram_e2e_greedy_correctness (vllm-project#19644) * [Kernel] Raise verbose error and consolidate `num_heads/num_kv_heads` divisibility check (vllm-project#19339) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Benchmark] Refactor benchmark script for fp8 & int8 (vllm-project#19627) Signed-off-by: yewentao256 <zhyanwentao@126.com> * Enable prefix caching with full cuda graphs (vllm-project#19617) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [CI/Build] Fix torch nightly CI dependencies part 2 (vllm-project#19589) * [Misc] Remove duplicate multiproc method setting for CPU platform (vllm-project#19649) Signed-off-by: Isotr0py <2037008807@qq.com> * [MISC] Remove unused variableds in C++ (vllm-project#19609) Signed-off-by: Lu Fang <lufang@fb.com> * [Bugfix][Core] Prefix caching causes incorrect outputs due to outdated ComputedBlocksTracker (vllm-project#18957) Signed-off-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Co-authored-by: 刘全 <quan.liu2@dbappsecurity.com.cn> * [Misc][Frontend] passthrough `bad_words` (vllm-project#19564) Signed-off-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> * [Misc] Fix skipped max-model-len validation when deriving max model length from tokenizer config (vllm-project#19660) Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> * [TPU] support attention head dim smaller than 128 (vllm-project#19620) Signed-off-by: Chengji Yao <chengjiyao@google.com> Co-authored-by: mgoin <mgoin64@gmail.com> * [MISC] typo fix (vllm-project#19672) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [CI] Add mteb testing for rerank models (vllm-project#19344) * [Docs] Move multiproc doc to v1 dir (vllm-project#19651) Signed-off-by: Russell Bryant <rbryant@redhat.com> * [Kernel] GGUF MMVQ kernel for multiple input vectors (vllm-project#18754) Signed-off-by: SzymonOzog <szymon.ozog@gmail.com> * [BugFix] Don't catch BaseException when dumping execute_model errors (vllm-project#19626) Signed-off-by: Nick Hill <nhill@redhat.com> * [DOC] Add reasoning capability to vLLM streamlit code (vllm-project#19557) * [Feature]:Allow for Granite MoE Hybrid models with _only_ shared experts. (vllm-project#19652) Signed-off-by: Shawn Tan <shawntan@ibm.com> * [Bugfix] Fix TP inference for Flex attention backend (vllm-project#19657) Signed-off-by: Isotr0py <2037008807@qq.com> * [MISC] bump huggingface_hub pkg to 0.33.0 (vllm-project#19547) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [Bugfix] fix missing 'finish_reason': null in streaming chat (vllm-project#19662) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Kernels] Use empty for modular MoE workspaces (vllm-project#19667) Signed-off-by: Bill Nell <bnell@redhat.com> * [Model] Add support for MiniMaxM1ForCausalLM (shares architecture with MiniMaxText01ForCausalLM) (vllm-project#19677) Signed-off-by: QscQ <qscqesze@gmail.com> * [V1] Change return type on get_multimodal_embeddings() (vllm-project#19446) Signed-off-by: Russell Bryant <rbryant@redhat.com> * fix Signed-off-by: Amog Kamsetty <amogkamsetty@gmail.com> * remove logging Signed-off-by: Amog Kamsetty <amogkamsetty@gmail.com> --------- Signed-off-by: raushan <raushan@huggingface.co> Signed-off-by: Lu Fang <lufang@fb.com> Signed-off-by: nicklucche <nlucches@redhat.com> Signed-off-by: googs1025 <googs1025@gmail.com> Signed-off-by: simon-mo <simon.mo@hey.com> Signed-off-by: reidliu41 <reid201711@gmail.com> Signed-off-by: Varun <vsundarr@redhat.com> Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: mgoin <mgoin64@gmail.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Signed-off-by: Chen Zhang <zhangch99@outlook.com> Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> Signed-off-by: Russell Bryant <rbryant@redhat.com> Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> Signed-off-by: calvin chen <120380290@qq.com> Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> Signed-off-by: Siyuan Liu <lsiyuan@google.com> Signed-off-by: Seiji Eicher <seiji@anyscale.com> Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> Signed-off-by: Jon Swenson <jmswen@gmail.com> Signed-off-by: Tyler Michael Smith <tysmith@redhat.com> Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> Signed-off-by: Nick Hill <nhill@redhat.com> Signed-off-by: Yang Wang <elainewy@meta.com> Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com> Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> Signed-off-by: Guillaume Calmettes <gcalmettes@scaleway.com> Signed-off-by: Patrick von Platen <patrick.v.platen@gmail.com> Signed-off-by: Chiyue Wei <chiyuew@nvidia.com> Signed-off-by: Povilas Kanapickas <povilas@radix.lt> Signed-off-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Signed-off-by: Jerry Zhang <jerryzh168@gmail.com> Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai> Signed-off-by: Chengji Yao <chengjiyao@google.com> Signed-off-by: Xu Song <xusong.vip@gmail.com> Signed-off-by: Aaron Pham <contact@aarnphm.xyz> Signed-off-by: Dipika Sikka <dipikasikka1@gmail.com> Signed-off-by: rzou <zou3519@gmail.com> Signed-off-by: Siqi Yan <siqi@meta.com> Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: Nishidha Panpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Signed-off-by: npanpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Chenyaaang <chenyangli@google.com> Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com> Signed-off-by: ElizaWszola <ewszola@redhat.com> Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com> Signed-off-by: Qiliang Cui <derrhein@gmail.com> Signed-off-by: Aaruni Aggarwal <aaruniagg@gmail.com> Signed-off-by: drisspg <drisspguessous@gmail.com> Signed-off-by: Lifan Shen <lifans@meta.com> Signed-off-by: pramkuma <Pramendra.Kumar@amd.com> Signed-off-by: luka <luka@neuralmagic.com> Signed-off-by: Richard Zou <zou3519@gmail.com> Signed-off-by: Xu Wenqing <xuwq1993@qq.com> Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Signed-off-by: yZhen <yZhen@fb.com> Signed-off-by: KsuParkhamchuk <k.parkhamchuk@gmail.com> Signed-off-by: cr7258 <chengzw258@163.com> Signed-off-by: Conroy Cheers <conroy@corncheese.org> Signed-off-by: windsonsea <haifeng.yao@daocloud.io> Signed-off-by: Yinghai Lu <yinghai@thinkingmachines.ai> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> Signed-off-by: liusiqian <liusiqian@tal.com> Signed-off-by: Pavani Majety <pmajety@nvidia.com> Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> Signed-off-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> Signed-off-by: Xiongfei Wei <isaacwxf23@gmail.com> Signed-off-by: wangli <wangli858794774@gmail.com> Signed-off-by: Anna Pendleton <pendleton@google.com> Signed-off-by: Tsai, Louie <louie.tsai@intel.com> Signed-off-by: Yunqiu Guo <guorachel@meta.com> Signed-off-by: jiang.li <jiang1.li@intel.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com> Signed-off-by: py-andy-c <pychen1017@gmail.com> Signed-off-by: niu_he <carlton2tang@gmail.com> Signed-off-by: Junhao Li <junhao@ubicloud.com> Signed-off-by: artetaout <lulala341@gmail.com> Signed-off-by: ximing.wxm <ximing.wxm@antgroup.com> Signed-off-by: Runzhen Wang <wangrunzhen@gmail.com> Signed-off-by: David Xia <david@davidxia.com> Signed-off-by: Bill Nell <bnell@redhat.com> Signed-off-by: Randall Smith <Randall.Smith@amd.com> Signed-off-by: Andy Xie <andy.xning@gmail.com> Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca> Signed-off-by: strutive07 <strutive07@gmail.com> Signed-off-by: 2niuhe <carlton2tang@gmail.com> Signed-off-by: NickLucche <nlucches@redhat.com> Signed-off-by: yewentao256 <zhyanwentao@126.com> Signed-off-by: mobicham <hicham@mobiuslabs.com> Signed-off-by: Luka Govedič <lgovedic@redhat.com> Signed-off-by: qizixi <qizixi@meta.com> Signed-off-by: Zerohertz <ohg3417@gmail.com> Signed-off-by: jiang1.li <jiang1.li@intel.com> Signed-off-by: Boyuan Feng <boyuan@meta.com> Signed-off-by: qingjun <qingjun@minimaxi.com> Signed-off-by: Yida Wu <yidawu@alumni.cmu.edu> Signed-off-by: Saheli Bhattacharjee <saheli@krai.ai> Signed-off-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Signed-off-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Signed-off-by: SzymonOzog <szymon.ozog@gmail.com> Signed-off-by: Shawn Tan <shawntan@ibm.com> Signed-off-by: QscQ <qscqesze@gmail.com> Signed-off-by: Amog Kamsetty <amogkamsetty@gmail.com> Co-authored-by: Raushan Turganbay <raushan.turganbay@alumni.nu.edu.kz> Co-authored-by: Lu Fang <30275821+houseroad@users.noreply.github.com> Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com> Co-authored-by: CYJiang <86391540+googs1025@users.noreply.github.com> Co-authored-by: Simon Mo <simon.mo@hey.com> Co-authored-by: SorenDreano <71752785+SorenDreano@users.noreply.github.com> Co-authored-by: Soren Dreano <soren@numind.ai> Co-authored-by: Reid <61492567+reidliu41@users.noreply.github.com> Co-authored-by: reidliu41 <reid201711@gmail.com> Co-authored-by: Varun Sundar Rabindranath <varunsundar08@gmail.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Yong Hoon Shin <48474650+sarckk@users.noreply.github.com> Co-authored-by: Michael Goin <mgoin64@gmail.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Yikun Jiang <yikun@apache.org> Co-authored-by: Chen Zhang <zhangch99@outlook.com> Co-authored-by: Ekagra Ranjan <3116519+ekagra-ranjan@users.noreply.github.com> Co-authored-by: Chauncey <chaunceyjiang@gmail.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> Co-authored-by: Yan Ru Pei <yanrpei@gmail.com> Co-authored-by: Jiaxin Shan <seedjeffwan@gmail.com> Co-authored-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Mark McLoughlin <markmc@redhat.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> Co-authored-by: Li, Jiang <jiang1.li@intel.com> Co-authored-by: Lukas Geiger <lukas.geiger94@gmail.com> Co-authored-by: Vadim Gimpelson <156319763+vadiklyutiy@users.noreply.github.com> Co-authored-by: Calvin Chen <45745657+calvin0327@users.noreply.github.com> Co-authored-by: Kaixi Hou <kaixih@nvidia.com> Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Co-authored-by: 汪志鹏 <wangzhipeng628@gmail.com> Co-authored-by: Siyuan Liu <lsiyuan@google.com> Co-authored-by: Seiji Eicher <58963096+eicherseiji@users.noreply.github.com> Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn> Co-authored-by: wang.yuqi <noooop@126.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk> Co-authored-by: Xu Wenqing <121550081+Xu-Wenqing@users.noreply.github.com> Co-authored-by: Lain <fusiyuan2000@hotmail.com> Co-authored-by: jmswen <jmswen@users.noreply.github.com> Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com> Co-authored-by: Kebe <mail@kebe7jun.com> Co-authored-by: Nick Hill <nhill@redhat.com> Co-authored-by: Yang Wang <elainewy@meta.com> Co-authored-by: Huy Do <huydhn@gmail.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com> Co-authored-by: 22quinn <33176974+22quinn@users.noreply.github.com> Co-authored-by: Guillaume Calmettes <gcalmettes@scaleway.com> Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> Co-authored-by: Chiyue Wei <92623189+dubcyfor3@users.noreply.github.com> Co-authored-by: Chiyue Wei <chiyuew@nvidia.com> Co-authored-by: Povilas Kanapickas <povilas@radix.lt> Co-authored-by: Dipika Sikka <dipikasikka1@gmail.com> Co-authored-by: Luis Vega <vegaluisjose@users.noreply.github.com> Co-authored-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Co-authored-by: Jerry Zhang <jerryzh168@gmail.com> Co-authored-by: Benjamin Chislett <benjamin.chislett@centml.ai> Co-authored-by: Chengji Yao <chengjiyao@google.com> Co-authored-by: Xu Song <xusong.vip@gmail.com> Co-authored-by: Aaron Pham <contact@aarnphm.xyz> Co-authored-by: Jinghui Zhang <jinghuizhang0804@gmail.com> Co-authored-by: jinghui <jinghui@fb.com> Co-authored-by: Richard Zou <zou3519@users.noreply.github.com> Co-authored-by: Siqi Yan <ysq0807@hotmail.com> Co-authored-by: Siqi Yan <siqi@meta.com> Co-authored-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Yu Guo <82124926+yuguo68@users.noreply.github.com> Co-authored-by: Nishidha <nishidha.panpaliya@partner.ibm.com> Co-authored-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Co-authored-by: Adolfo Victoria <adolfokarim@gmail.com> Co-authored-by: Adolfo Victoria <adovi@meta.com> Co-authored-by: Chenyaaang <42742451+Chenyaaang@users.noreply.github.com> Co-authored-by: Alexei-V-Ivanov-AMD <156011006+Alexei-V-Ivanov-AMD@users.noreply.github.com> Co-authored-by: ElizaWszola <ewszola@redhat.com> Co-authored-by: QiliangCui <derrhein@gmail.com> Co-authored-by: Aaruni Aggarwal <47731267+AaruniAggarwal@users.noreply.github.com> Co-authored-by: Driss Guessous <32754868+drisspg@users.noreply.github.com> Co-authored-by: Lifans <draftbks@gmail.com> Co-authored-by: pramenku <7664080+pramenku@users.noreply.github.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Co-authored-by: Akash kaothalkar <61960177+Akashcodes732@users.noreply.github.com> Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Co-authored-by: jennyyyyzhen <47012288+jennyyyyzhen@users.noreply.github.com> Co-authored-by: yZhen <yZhen@fb.com> Co-authored-by: Kseniya Parkhamchuk <43078183+KsuParkhamchuk@users.noreply.github.com> Co-authored-by: Se7en <chengzw258@163.com> Co-authored-by: Conroy Cheers <conroy@corncheese.org> Co-authored-by: Michael Yao <haifeng.yao@daocloud.io> Co-authored-by: Yinghai Lu <yinghai@thinkingmachines.ai> Co-authored-by: Kyle Sayers <kylesayrs@gmail.com> Co-authored-by: liusiqian-tal <141730978+liusiqian-tal@users.noreply.github.com> Co-authored-by: Pavani Majety <pmajety@nvidia.com> Co-authored-by: Ye (Charlotte) Qi <yeq@meta.com> Co-authored-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> Co-authored-by: XiongfeiWei <isaacwxf23@gmail.com> Co-authored-by: Li Wang <wangli858794774@gmail.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: Anna Pendleton <pendleton@google.com> Co-authored-by: Louie Tsai <louie.tsai@intel.com> Co-authored-by: Li, Jiang <bigpyj64@gmail.com> Co-authored-by: Rachel Guo <35738743+YUNQIUGUO@users.noreply.github.com> Co-authored-by: youkaichao <youkaichao@gmail.com> Co-authored-by: Isotr0py <2037008807@qq.com> Co-authored-by: Gregory Shtrasberg <156009573+gshtras@users.noreply.github.com> Co-authored-by: py-andy-c <37168711+py-andy-c@users.noreply.github.com> Co-authored-by: niu_he <carlton2tang@gmail.com> Co-authored-by: Junhao Li <junhao@ubicloud.com> Co-authored-by: leopardracer <136604165+leopardracer@users.noreply.github.com> Co-authored-by: artetaout <128046886+artetaout@users.noreply.github.com> Co-authored-by: Ximingwang-09 <72070413+Ximingwang-09@users.noreply.github.com> Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com> Co-authored-by: runzhen <wangrunzhen@gmail.com> Co-authored-by: David Xia <david@davidxia.com> Co-authored-by: bnellnm <49004751+bnellnm@users.noreply.github.com> Co-authored-by: rasmith <Randall.Smith@amd.com> Co-authored-by: Ning Xie <andy.xning@gmail.com> Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca> Co-authored-by: wonjun Jang <strutive07@gmail.com> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Co-authored-by: mobicham <37179323+mobicham@users.noreply.github.com> Co-authored-by: Sage Moore <sage@neuralmagic.com> Co-authored-by: kourosh hakhamaneshi <31483498+kouroshHakha@users.noreply.github.com> Co-authored-by: qizixi <22851944+zixi-qi@users.noreply.github.com> Co-authored-by: Hyogeun Oh (오효근) <ohg3417@gmail.com> Co-authored-by: Boyuan Feng <fby.1994@gmail.com> Co-authored-by: qscqesze <qingjun@minimaxi.com> Co-authored-by: Concurrensee <yida.wu@amd.com> Co-authored-by: Saheli Bhattacharjee <47847054+sahelib25@users.noreply.github.com> Co-authored-by: jiahanc <173873397+jiahanc@users.noreply.github.com> Co-authored-by: Konrad Zawora <kzawora@habana.ai> Co-authored-by: maobaolong <baoloongmao@tencent.com> Co-authored-by: Ilya Markov <markovilya197@gmail.com> Co-authored-by: quanliu <33453350+quanliu1991@users.noreply.github.com> Co-authored-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Co-authored-by: Francesco Bertolotti <f14.bertolotti@gmail.com> Co-authored-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Szymon Ożóg <58388001+SzymonOzog@users.noreply.github.com> Co-authored-by: Navanit Dubey <98005188+Navanit-git@users.noreply.github.com> Co-authored-by: Shawn Tan <shawntan@ibm.com> Co-authored-by: qscqesze <qscqesze@gmail.com>
amogkam
added a commit
to character-tech/vllm
that referenced
this pull request
Jun 16, 2025
* [Bugfix] disable processor cache (vllm-project#19068) Signed-off-by: raushan <raushan@huggingface.co> * [Doc] Improve the Pull Request template with key components (vllm-project#19086) Signed-off-by: Lu Fang <lufang@fb.com> * [Misc] Add missing `_Backend` enums (vllm-project#19081) Signed-off-by: nicklucche <nlucches@redhat.com> * [Misc] fix: add miss best_of param validation (vllm-project#18555) Signed-off-by: googs1025 <googs1025@gmail.com> * [Misc] Add SPDX-FileCopyrightText (vllm-project#19100) Signed-off-by: simon-mo <simon.mo@hey.com> * [Doc] Readme standardization (vllm-project#18695) Co-authored-by: Soren Dreano <soren@numind.ai> * [doc] update docker version (vllm-project#19074) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Kernel] DeepEP dispatch-combine kernel integration (vllm-project#18434) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * [V1] Support cross-layer KV sharing (vllm-project#18212) Signed-off-by: Yong Hoon Shin <yhshin@meta.com> * [Perf] Tune `scaled_fp8_quant` by increasing vectorization (vllm-project#18844) Signed-off-by: mgoin <mgoin64@gmail.com> * Fix interaction between `Optional` and `Annotated` in CLI typing (vllm-project#19093) Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Yikun Jiang <yikun@apache.org> * [v1] Re-init input batch for multiple kv cache groups (vllm-project#18654) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [V1][Spec Decode][Ngram] 1.35x gain -> 1.95x gain on InstructCoder with prompt fix (vllm-project#18971) * [Bugfix] get_num_blocks_to_allocate with null_block (vllm-project#19031) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [Bugfix]: Fix the incompatibility issue with tool_choice 'required' when Thinking is enabled (vllm-project#19075) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Bugfix][P/D] Fix Prefix Cache Bug (vllm-project#18411) Signed-off-by: nicklucche <nlucches@redhat.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> * [Bugfix] Max concurrency estimation and check_enough_kv_cache_memory for models with sliding window layers (vllm-project#19029) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * feat: add data parallel rank to KVEventBatch (vllm-project#18925) * [Misc] Fix path and python alias errors in disagg_prefill exmaples (vllm-project#18919) * [Docs] Add developer doc about CI failures (vllm-project#18782) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Mark McLoughlin <markmc@redhat.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> * [CPU] V1 support for the CPU backend (vllm-project#16441) * [Core] Cast multimodal input in hf processor (vllm-project#18862) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * [KERNEL] Sampler. CUDA kernel for applying repetition penalty (vllm-project#18437) * [Cleanup][v1]:remote guided-decoding-backend for example (vllm-project#19059) Signed-off-by: calvin chen <120380290@qq.com> * [NVIDIA] Add Cutlass MLA backend (vllm-project#17625) * [Bugfix] Fix FA3 full cuda graph correctness (vllm-project#19106) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * Fix vllm-project#19130 (vllm-project#19132) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * [TPU] Skip hanging tests (vllm-project#19115) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * Fix ValueError: Missing value for tag key(s): model_name,engine. (vllm-project#19113) Signed-off-by: Seiji Eicher <seiji@anyscale.com> * [Misc] Add packages for benchmark as extra dependency (vllm-project#19089) Signed-off-by: Isotr0py <2037008807@qq.com> * Improve the output precision of embedding models (vllm-project#19092) * [CI/Build][Bugfix] Ensure compatibility with transformers 4.52 (vllm-project#18678) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Add DeepSeek-R1-0528 function call chat template (vllm-project#18874) Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> * Sm100 blockwise fp8 swap ab (vllm-project#18564) * [Doc] Update V1 Guide for embedding models (vllm-project#19141) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Allow AsyncLLMEngine.generate to target a specific DP rank (vllm-project#19102) Signed-off-by: Jon Swenson <jmswen@gmail.com> * [Bugfix][EP+DP] Fix internode check (vllm-project#19112) Signed-off-by: Tyler Michael Smith <tysmith@redhat.com> * [Perf] Tunings for SM100 FP8 CUTLASS kernel (vllm-project#18778) Signed-off-by: mgoin <mgoin64@gmail.com> * [TPU] Update dynamo dump file name in compilation test (vllm-project#19108) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * [Bugfix] fix v1 cpu worker fails on macOS (vllm-project#19121) * [Kernel] Integrate batched/masked deepgemm kernel (vllm-project#19111) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun <vsundarr@redhat.com> * [Misc] refactor: simplify EngineCoreClient.make_async_mp_client in AsyncLLM (vllm-project#18817) Signed-off-by: googs1025 <googs1025@gmail.com> * [P/D] Heterogeneous TP (vllm-project#18833) Signed-off-by: nicklucche <nlucches@redhat.com> * [doc] small fix (vllm-project#19167) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix][Nixl] Fix full prefix cache hit bug (vllm-project#18632) Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> Signed-off-by: Nick Hill <nhill@redhat.com> Co-authored-by: Nick Hill <nhill@redhat.com> * [Bugfix] Fix port handling in make_zmq_path (vllm-project#19117) * [Torch Nightly]add missing dependency (vllm-project#18770) Signed-off-by: Yang Wang <elainewy@meta.com> * Handle non-serializable objects when dumping benchmark results (vllm-project#19114) * [BugFix][Minor] Fix full cuda graph bug when max_num_seqs < 512 (vllm-project#19171) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [Bugfix]: Fix the incompatibility issue with stream when Thinking is disabled (vllm-project#19135) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Build] Annotate wheel and container path for release workflow (vllm-project#19162) Signed-off-by: simon-mo <simon.mo@hey.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * [Misc] Remove unnecessary fallback to prefill-decode attention (vllm-project#19138) Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com> * [Misc] Do not override NCCL_CUMEM_ENABLE if set explicitly (vllm-project#19105) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Frontend] improve vllm run-batch --help display (vllm-project#19187) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix] properly catch PIL-related errors for vision models when incorrect data urls are provided (vllm-project#19202) Signed-off-by: Guillaume Calmettes <gcalmettes@scaleway.com> * [mistral_common] Add v11 tokenizer (vllm-project#19193) Signed-off-by: Patrick von Platen <patrick.v.platen@gmail.com> * Add H20-3e fused MoE kernel tuning configs for DeepSeek-R1/V3 (vllm-project#19205) * [Hardware][NVIDIA] FP4 MoE kernel optimization (vllm-project#19110) Signed-off-by: Chiyue Wei <chiyuew@nvidia.com> Co-authored-by: Chiyue Wei <chiyuew@nvidia.com> * [MISC][Bugfix] Use less CPU when message queue has been empty for some time (vllm-project#16226) Signed-off-by: Povilas Kanapickas <povilas@radix.lt> * [P/D][NixlConnector] Enable FlashInfer backend (vllm-project#19090) * [Quantization] Skip Fp4 Test for `compressed-tensors` (vllm-project#19217) * [V1] Use FlashInfer by default on Blackwell GPUs (vllm-project#19118) * [Model] NemotronH support (vllm-project#18863) Signed-off-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Co-authored-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> * Fix AOPerModuleConfig name changes (vllm-project#18869) Signed-off-by: Jerry Zhang <jerryzh168@gmail.com> * [Bugfix] Fix EAGLE vocab embedding construction for Llama 70B (vllm-project#19033) Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai> * [v1] Hybrid Memory Allocator (vllm-project#17996) Signed-off-by: Chen Zhang <zhangch99@outlook.com> * [TPU] update torch_xla pin (vllm-project#19231) Signed-off-by: Chengji Yao <chengjiyao@google.com> * Support allowed_token_ids in ChatCompletionRequest (vllm-project#19143) Signed-off-by: Xu Song <xusong.vip@gmail.com> * [Chore] update CODEOWNERS (vllm-project#19247) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [v1][P/D] Fix a edge case in kv cache schedule (vllm-project#19182) Co-authored-by: jinghui <jinghui@fb.com> * [TPU] fix kv cache dtype in model runner (vllm-project#19244) Signed-off-by: Chengji Yao <chengjiyao@google.com> * [Quantization] Bump compressed-tensors version; update NVFP4A16 test model (vllm-project#19224) Signed-off-by: Dipika Sikka <dipikasikka1@gmail.com> * [Docs] Improve V1 KVConnector interface documentation (vllm-project#19172) Signed-off-by: Nick Hill <nhill@redhat.com> * Fix CompilationConfig repr (vllm-project#19091) Signed-off-by: rzou <zou3519@gmail.com> * Unit Test for run_dp_sharded_vision_model (vllm-project#19103) Signed-off-by: Siqi Yan <siqi@meta.com> Co-authored-by: Siqi Yan <siqi@meta.com> * [Model] Optimize nemotron_h implementation (vllm-project#19249) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Core] Raise when non-multi-instance DP clients target a DP rank (vllm-project#19227) Signed-off-by: Jon Swenson <jmswen@gmail.com> * improve logits bias (vllm-project#19041) * Fixed ppc build when it runs on non-RHEL based linux distros (vllm-project#18422) Signed-off-by: Nishidha Panpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Signed-off-by: npanpaliya <nishidha.panpaliya@partner.ibm.com> Co-authored-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> * [BugFix] Fix MultiConnector test after HMA changes (vllm-project#19291) Signed-off-by: Nick Hill <nhill@redhat.com> * [Bugfix][Core] Update cancellation logic in `generate()` to handle Generator exits (vllm-project#19225) Co-authored-by: Adolfo Victoria <adovi@meta.com> * [Core] Fix abrupt request abort (vllm-project#18485) Signed-off-by: nicklucche <nlucches@redhat.com> Signed-off-by: Nick Hill <nhill@redhat.com> Co-authored-by: Nick Hill <nhill@redhat.com> * [BugFix] Fix tpu_model_runner block_id concatenation (vllm-project#19228) Signed-off-by: Nick Hill <nhill@redhat.com> * [Misc][Tools][Benchmark] Fix and improve auto tune script (vllm-project#19163) Signed-off-by: Chenyaaang <chenyangli@google.com> * [Build][ROCm] Update Dockerfile.rocm (vllm-project#19296) Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com> * [Easy][Test] Simplify test_function_tool_use with multiple parametrizes (vllm-project#19269) Signed-off-by: Lu Fang <lufang@fb.com> * [Kernel] Integrate CUTLASS MoE kernel with PPLX (vllm-project#18762) Signed-off-by: ElizaWszola <ewszola@redhat.com> Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com> Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com> * [TPU][Test] Add script to run benchmark on TPU for buildkite (vllm-project#19039) Signed-off-by: Qiliang Cui <derrhein@gmail.com> * [CI][PowerPC] Use a more appropriate way to select testcase in tests/models/language/pooling/test_embedding.py (vllm-project#19253) Signed-off-by: Aaruni Aggarwal <aaruniagg@gmail.com> * Add FlexAttention to V1 (vllm-project#16078) Signed-off-by: drisspg <drisspguessous@gmail.com> * [Misc] refactor context extension (vllm-project#19246) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [CI/Build] Improve Llama GGUF test robustness (vllm-project#19287) Signed-off-by: Isotr0py <2037008807@qq.com> * [Nit][Benchmark]Fix example in benchmark_serving_structured_output.py (vllm-project#19311) Signed-off-by: Lifan Shen <lifans@meta.com> * [AMD] Update compatible packaging version (vllm-project#19309) Signed-off-by: pramkuma <Pramendra.Kumar@amd.com> * [BugFix][V1] Fix memory profiling bug (vllm-project#18974) Signed-off-by: luka <luka@neuralmagic.com> * [Bugfix]: Fix TypeError: 'float' object cannot be interpreted as an integer (vllm-project#19283) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Bugfix] Re-enable use_cudagraph in vLLM v1 (vllm-project#19299) Signed-off-by: Richard Zou <zou3519@gmail.com> * [Misc] Change tests/compile to use VLLM_V1 by default (vllm-project#19302) Signed-off-by: rzou <zou3519@gmail.com> * Add H20-3e fused MoE kernel tuning configs for Qwen3-235B-A22B (vllm-project#19315) Signed-off-by: Xu Wenqing <xuwq1993@qq.com> * [Hardware][POWER] Add IBM POWER11 Support to CPU Extension Detection (vllm-project#19082) Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> * [Quantization] Add compressed-tensors NVFP4 support (vllm-project#18312) * [Multi Modal] Add an env var for message queue max chunk bytes (vllm-project#19242) Signed-off-by: yZhen <yZhen@fb.com> Co-authored-by: yZhen <yZhen@fb.com> * [Bugfix] model_max_length should consider max_model_len in tokenizer_config (vllm-project#19201) * [Deprecation] Remove `inputs` arg fallback in Engine classes (vllm-project#18799) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Misc] Add documentation update reminder to PR template (vllm-project#19289) Signed-off-by: Isotr0py <2037008807@qq.com> * [Frontend] Remove unreachable code from llm.py (vllm-project#19288) Signed-off-by: KsuParkhamchuk <k.parkhamchuk@gmail.com> * [Misc] Cleanup compilation tests (vllm-project#19343) Signed-off-by: rzou <zou3519@gmail.com> * [doc] improve ci doc (vllm-project#19307) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Doc] Fix description in the Automatic Prefix Caching design doc (vllm-project#19333) Signed-off-by: cr7258 <chengzw258@163.com> * [CI/Build] Fix LoRA test (vllm-project#19350) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Fix] Allow kernel compilation for CUDA capability 8.7 (vllm-project#19328) Signed-off-by: Conroy Cheers <conroy@corncheese.org> * [CI] Introduce rules for llama auto-label (vllm-project#19323) Signed-off-by: Lu Fang <lufang@fb.com> * [Docs] Fix a bullet list in usage/security.md (vllm-project#19358) Signed-off-by: windsonsea <haifeng.yao@daocloud.io> * [full_graph] Fix query_start_loc padding (vllm-project#19321) Signed-off-by: Yinghai Lu <yinghai@thinkingmachines.ai> * [v1] Add fp32 support to v1 engine through flex attn (vllm-project#19319) Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> * [Misc] Fixes and Optimizations for DeepEP + DeepGEMM combination. (vllm-project#19298) Signed-off-by: Varun <vsundarr@redhat.com> Co-authored-by: Varun <vsundarr@redhat.com> * [Bugfix][Core] Prevent token lengths exceeding `max_model_len` in V0 (vllm-project#19348) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Quantization] Bump compressed-tensors version (vllm-project#19295) Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> * [Frontend] Make TIMEOUT_KEEP_ALIVE configurable through env var (vllm-project#18472) Signed-off-by: liusiqian <liusiqian@tal.com> * [TPU]Fix KV cache sharing tests (vllm-project#19371) * [HOT-FIX] Add `kv_sharing_target_layer_name` argument to cutlass_mla backend (vllm-project#19374) Signed-off-by: Pavani Majety <pmajety@nvidia.com> * [Misc] Fix a config typo in disable_hybrid_kv_cache_manager configuration (vllm-project#19383) Signed-off-by: Siyuan Liu <lsiyuan@google.com> * [V1] Reuse V0's memory_profiling util for gpu worker memory profiling (vllm-project#19312) Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> * [Bugfix] Fix benchmark_moe.py (vllm-project#19016) Signed-off-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> * Use xla flag to improve the quantized model performance (vllm-project#19303) Signed-off-by: Xiongfei Wei <isaacwxf23@gmail.com> * Fix docs/mkdocs/hooks/remove_announcement.py (vllm-project#19382) * [Frontend] Add tqdm_leave_pbar to control progress bar visibility (vllm-project#19357) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Core] Use tuple for kv cache group block ids (vllm-project#19175) Signed-off-by: Nick Hill <nhill@redhat.com> * [Bugfix] Fix modelscope token passed in (vllm-project#19389) Signed-off-by: wangli <wangli858794774@gmail.com> Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Jee Jee Li <pandaleefree@gmail.com> * [Core] Batch multi modal input using pinned memory (vllm-project#19169) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * Add security warning to bug report template (vllm-project#19365) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * [Misc] refactor neuron_multimodal and profiling (vllm-project#19397) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * Add clear documentation around the impact of debugging flag (vllm-project#19369) Signed-off-by: Anna Pendleton <pendleton@google.com> * Automatically bind CPU OMP Threads of a rank to CPU ids of a NUMA node. (vllm-project#17930) Signed-off-by: Tsai, Louie <louie.tsai@intel.com> Co-authored-by: Li, Jiang <bigpyj64@gmail.com> * Revert "[v1] Add fp32 support to v1 engine through flex attn" (vllm-project#19404) * [BugFix][FlashInfer] Fix attention backend interface mismatch with unexpected keyword `use_irope` (vllm-project#19134) Signed-off-by: Yunqiu Guo <guorachel@meta.com> * [BugFix][CPU] Fix CPU CI by ignore collecting test_pixtral (vllm-project#19411) Signed-off-by: jiang.li <jiang1.li@intel.com> * Simplify ep kernels installation (vllm-project#19412) Signed-off-by: youkaichao <youkaichao@gmail.com> * [Misc] Slight improvement of the BNB (vllm-project#19418) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Isotr0py <2037008807@qq.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * [Docs] Note that alternative structured output backends are supported (vllm-project#19426) Signed-off-by: Russell Bryant <rbryant@redhat.com> * [ROCm][V1] Adding ROCm to the list of plaforms using V1 by default (vllm-project#19440) Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com> * [Model] use AutoWeightsLoader for commandr (vllm-project#19399) Signed-off-by: py-andy-c <pychen1017@gmail.com> * Add H20-3e fused MoE kernel tuning configs for Qwen3-235B-A22B-FP8 (vllm-project#19401) Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> * [BugFix] Allow use_cudagraph to work with dynamic VLLM_USE_V1 (vllm-project#19390) Signed-off-by: rzou <zou3519@gmail.com> * [New Model]: Support Qwen3 Embedding & Reranker (vllm-project#19260) * [BugFix] Fix docker build cpu-dev image error (vllm-project#19394) Signed-off-by: niu_he <carlton2tang@gmail.com> * Fix test_max_model_len in tests/entrypoints/llm/test_generate.py (vllm-project#19451) Signed-off-by: Lu Fang <lufang@fb.com> * [CI] Disable failing GGUF model test (vllm-project#19454) Signed-off-by: mgoin <mgoin64@gmail.com> * [Misc] Remove unused `MultiModalHasher.hash_prompt_mm_data` (vllm-project#19422) Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> * Add fused MOE config for Qwen3 30B A3B on B200 (vllm-project#19455) Signed-off-by: Junhao Li <junhao@ubicloud.com> * Fix Typo in Documentation and Function Name (vllm-project#19442) * [ROCm] Add rules to automatically label ROCm related PRs (vllm-project#19405) Signed-off-by: Lu Fang <lufang@fb.com> * [Kernel] Support deep_gemm for linear methods (vllm-project#19085) Signed-off-by: artetaout <lulala341@gmail.com> * [Doc] Update V1 User Guide for Hardware and Models (vllm-project#19474) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Doc] Fix quantization link titles (vllm-project#19478) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Doc] Support "important" and "announcement" admonitions (vllm-project#19479) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [Misc] Reduce warning message introduced in env_override (vllm-project#19476) Signed-off-by: Lu Fang <lufang@fb.com> * Support non-string values in JSON keys from CLI (vllm-project#19471) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * Add cache to cuda get_device_capability (vllm-project#19436) Signed-off-by: mgoin <mgoin64@gmail.com> * Fix some typo (vllm-project#19475) Signed-off-by: ximing.wxm <ximing.wxm@antgroup.com> Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com> * Support no privileged mode on CPU for docker and kubernetes deployments (vllm-project#19241) Signed-off-by: Tsai, Louie <louie.tsai@intel.com> * [Bugfix] Update the example code, make it work with the latest lmcache (vllm-project#19453) Signed-off-by: Runzhen Wang <wangrunzhen@gmail.com> * [CI] Update FlashInfer to 0.2.6.post1 (vllm-project#19297) Signed-off-by: mgoin <mgoin64@gmail.com> * [doc] fix "Other AI accelerators" getting started page (vllm-project#19457) Signed-off-by: David Xia <david@davidxia.com> * [Misc] Fix misleading ROCm warning (vllm-project#19486) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Docs] Remove WIP features in V1 guide (vllm-project#19498) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [Kernels] Add activation chunking logic to FusedMoEModularKernel (vllm-project#19168) Signed-off-by: Bill Nell <bnell@redhat.com> * [AMD] [Quantization] Add override flag for attention dtype instead of using kv_cache_dtype trigger (vllm-project#17331) Signed-off-by: Randall Smith <Randall.Smith@amd.com> * [UX] Add Feedback During CUDAGraph Capture (vllm-project#19501) Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> * [CI/Build] Fix torch nightly CI dependencies (vllm-project#19505) Signed-off-by: Richard Zou <zou3519@gmail.com> * [CI] change spell checker from codespell to typos (vllm-project#18711) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [BugFix] Force registration of w8a8_block_fp8_matmul_deepgemm via lazy import (vllm-project#19514) Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * Add Triton Fused MoE kernel config for E=16 on B200 (vllm-project#19518) Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca> * [Frontend] Improve error message in tool_choice validation (vllm-project#19239) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [BugFix] Work-around incremental detokenization edge case error (vllm-project#19449) Signed-off-by: Nick Hill <nhill@redhat.com> * [BugFix] Handle missing sep_token for Qwen3-Reranker in Score API (vllm-project#19522) Signed-off-by: strutive07 <strutive07@gmail.com> * [AMD][Kernel][BugFix] fix test_rocm_compressed_tensors_w8a8 for rocm (vllm-project#19509) Signed-off-by: Randall Smith <Randall.Smith@amd.com> * Fix typo (vllm-project#19525) Signed-off-by: 2niuhe <carlton2tang@gmail.com> * [Security] Prevent new imports of (cloud)pickle (vllm-project#18018) Signed-off-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> * [Bugfix][V1] Allow manual FlashAttention for Blackwell (vllm-project#19492) Signed-off-by: mgoin <mgoin64@gmail.com> * [Bugfix] Respect num-gpu-blocks-override in v1 (vllm-project#19503) Signed-off-by: Jon Swenson <jmswen@gmail.com> * [Quantization] Improve AWQ logic (vllm-project#19431) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> * [Doc] Add V1 column to supported models list (vllm-project#19523) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> * [V1][NixlConnector] Drop `num_blocks` check (vllm-project#19532) Signed-off-by: NickLucche <nlucches@redhat.com> * [Perf] Vectorize static / dynamic INT8 quant kernels (vllm-project#19233) Signed-off-by: yewentao256 <zhyanwentao@126.com> * Fix TorchAOConfig skip layers (vllm-project#19265) Signed-off-by: mobicham <hicham@mobiuslabs.com> * [torch.compile][ROCm] Fuse quantization onto attention using a torch.compile pass (vllm-project#16756) Signed-off-by: Luka Govedič <lgovedic@redhat.com> Co-authored-by: Sage Moore <sage@neuralmagic.com> * [doc] Make top navigation sticky (vllm-project#19540) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Spec Decode][Benchmark] Generalize spec decode offline benchmark to more methods and datasets (vllm-project#18847) * [Misc] Turn MOE_DP_CHUNK_SIZE into an env var (vllm-project#19506) * [Bugfix] Enforce contiguous input for dynamic_per_token FP8/INT8 quant (vllm-project#19452) Signed-off-by: mgoin <mgoin64@gmail.com> * [Doc] Unify structured outputs examples (vllm-project#18196) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [V1] Resolve failed concurrent structured output requests (vllm-project#19565) Signed-off-by: Russell Bryant <rbryant@redhat.com> * Revert "[Build/CI] Add tracing deps to vllm container image (vllm-project#15224)" (vllm-project#19378) * [BugFix] : Fix Batched DeepGemm Experts (vllm-project#19515) Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> * [Bugfix] Fix EAGLE vocab embedding for multimodal target model (vllm-project#19570) Signed-off-by: qizixi <qizixi@meta.com> * [Doc] uses absolute links for structured outputs (vllm-project#19582) Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * [doc] fix incorrect link (vllm-project#19586) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Misc] Correct broken docs link (vllm-project#19553) Signed-off-by: Zerohertz <ohg3417@gmail.com> * [CPU] Refine default config for the CPU backend (vllm-project#19539) Signed-off-by: jiang1.li <jiang1.li@intel.com> * [Fix] bump mistral common to support magistral (vllm-project#19533) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * [Fix] The zip function in Python 3.9 does not have the strict argument (vllm-project#19549) Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> * use base version for version comparison (vllm-project#19587) Signed-off-by: Boyuan Feng <boyuan@meta.com> * [torch.compile] reorganize the cache directory to support compiling multiple models (vllm-project#19064) Signed-off-by: youkaichao <youkaichao@gmail.com> * [BugFix] Honor `enable_caching` in connector-delayed kvcache load case (vllm-project#19435) Signed-off-by: Nick Hill <nhill@redhat.com> * [Model] Fix minimax model cache & lm_head precision (vllm-project#19592) Signed-off-by: qingjun <qingjun@minimaxi.com> * [Refactor] Remove unused variables in `moe_permute_unpermute_kernel.inl` (vllm-project#19573) Signed-off-by: yewentao256 <zhyanwentao@126.com> * [doc][mkdocs] fix the duplicate Supported features sections in GPU docs (vllm-project#19606) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [CUDA] Enable full cudagraph for FlashMLA (vllm-project#18581) Signed-off-by: luka <luka@neuralmagic.com> * [Doc] Add troubleshooting section to k8s deployment (vllm-project#19377) Signed-off-by: Anna Pendleton <pendleton@google.com> * [torch.compile] Use custom ops when use_inductor=False (vllm-project#19618) * Adding "AMD: Multi-step Tests" to amdproduction. (vllm-project#19508) Signed-off-by: Yida Wu <yidawu@alumni.cmu.edu> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> * [BugFix] Fix DP Coordinator incorrect debug log message (vllm-project#19624) Signed-off-by: Nick Hill <nhill@redhat.com> * [V1][Metrics] Deprecate metrics with gpu_ prefix for non GPU specific metrics. (vllm-project#18354) Signed-off-by: Saheli Bhattacharjee <saheli@krai.ai> * [Bugfix] Fix the speculative decoding test by setting the target dtype (vllm-project#19633) * [Misc] Modularize CLI Argument Parsing in Benchmark Scripts (vllm-project#19593) Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com> * [Bugfix] Fix auto dtype casting for BatchFeature (vllm-project#19316) Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> * [Hardware][NVIDIA][kernel] Fp4 MOE quant kernel optimization (vllm-project#19500) * Only build CUTLASS MoE kernels on Hopper (vllm-project#19648) * [Bugfix] Don't attempt to use triton if no driver is active (vllm-project#19561) * [Fix] Convert kv_transfer_config from dict to KVTransferConfig (vllm-project#19262) * [Perf] Further tunings for SM100 FP8 CUTLASS kernel (vllm-project#19566) * [Bugfix][2/n] Fix speculative decoding CI - Fix test_ngram_e2e_greedy_correctness (vllm-project#19644) * [Kernel] Raise verbose error and consolidate `num_heads/num_kv_heads` divisibility check (vllm-project#19339) Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> * [Benchmark] Refactor benchmark script for fp8 & int8 (vllm-project#19627) Signed-off-by: yewentao256 <zhyanwentao@126.com> * Enable prefix caching with full cuda graphs (vllm-project#19617) Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> * [CI/Build] Fix torch nightly CI dependencies part 2 (vllm-project#19589) * [Misc] Remove duplicate multiproc method setting for CPU platform (vllm-project#19649) Signed-off-by: Isotr0py <2037008807@qq.com> * [MISC] Remove unused variableds in C++ (vllm-project#19609) Signed-off-by: Lu Fang <lufang@fb.com> * [Bugfix][Core] Prefix caching causes incorrect outputs due to outdated ComputedBlocksTracker (vllm-project#18957) Signed-off-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Co-authored-by: 刘全 <quan.liu2@dbappsecurity.com.cn> * [Misc][Frontend] passthrough `bad_words` (vllm-project#19564) Signed-off-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> * [Misc] Fix skipped max-model-len validation when deriving max model length from tokenizer config (vllm-project#19660) Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> * [TPU] support attention head dim smaller than 128 (vllm-project#19620) Signed-off-by: Chengji Yao <chengjiyao@google.com> Co-authored-by: mgoin <mgoin64@gmail.com> * [MISC] typo fix (vllm-project#19672) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [CI] Add mteb testing for rerank models (vllm-project#19344) * [Docs] Move multiproc doc to v1 dir (vllm-project#19651) Signed-off-by: Russell Bryant <rbryant@redhat.com> * [Kernel] GGUF MMVQ kernel for multiple input vectors (vllm-project#18754) Signed-off-by: SzymonOzog <szymon.ozog@gmail.com> * [BugFix] Don't catch BaseException when dumping execute_model errors (vllm-project#19626) Signed-off-by: Nick Hill <nhill@redhat.com> * [DOC] Add reasoning capability to vLLM streamlit code (vllm-project#19557) * [Feature]:Allow for Granite MoE Hybrid models with _only_ shared experts. (vllm-project#19652) Signed-off-by: Shawn Tan <shawntan@ibm.com> * [Bugfix] Fix TP inference for Flex attention backend (vllm-project#19657) Signed-off-by: Isotr0py <2037008807@qq.com> * [MISC] bump huggingface_hub pkg to 0.33.0 (vllm-project#19547) Signed-off-by: Andy Xie <andy.xning@gmail.com> * [Bugfix] fix missing 'finish_reason': null in streaming chat (vllm-project#19662) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> * [Kernels] Use empty for modular MoE workspaces (vllm-project#19667) Signed-off-by: Bill Nell <bnell@redhat.com> * [Model] Add support for MiniMaxM1ForCausalLM (shares architecture with MiniMaxText01ForCausalLM) (vllm-project#19677) Signed-off-by: QscQ <qscqesze@gmail.com> * [V1] Change return type on get_multimodal_embeddings() (vllm-project#19446) Signed-off-by: Russell Bryant <rbryant@redhat.com> * fix Signed-off-by: Amog Kamsetty <amogkamsetty@gmail.com> --------- Signed-off-by: raushan <raushan@huggingface.co> Signed-off-by: Lu Fang <lufang@fb.com> Signed-off-by: nicklucche <nlucches@redhat.com> Signed-off-by: googs1025 <googs1025@gmail.com> Signed-off-by: simon-mo <simon.mo@hey.com> Signed-off-by: reidliu41 <reid201711@gmail.com> Signed-off-by: Varun <vsundarr@redhat.com> Signed-off-by: Yong Hoon Shin <yhshin@meta.com> Signed-off-by: mgoin <mgoin64@gmail.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Signed-off-by: Chen Zhang <zhangch99@outlook.com> Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> Signed-off-by: Russell Bryant <rbryant@redhat.com> Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com> Signed-off-by: calvin chen <120380290@qq.com> Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com> Signed-off-by: Siyuan Liu <lsiyuan@google.com> Signed-off-by: Seiji Eicher <seiji@anyscale.com> Signed-off-by: Isotr0py <2037008807@qq.com> Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: 许文卿 <xwq391974@alibaba-inc.com> Signed-off-by: Jon Swenson <jmswen@gmail.com> Signed-off-by: Tyler Michael Smith <tysmith@redhat.com> Signed-off-by: rshaw@neuralmagic.com <robertgshaw2@gmail.com> Signed-off-by: Nick Hill <nhill@redhat.com> Signed-off-by: Yang Wang <elainewy@meta.com> Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com> Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com> Signed-off-by: Guillaume Calmettes <gcalmettes@scaleway.com> Signed-off-by: Patrick von Platen <patrick.v.platen@gmail.com> Signed-off-by: Chiyue Wei <chiyuew@nvidia.com> Signed-off-by: Povilas Kanapickas <povilas@radix.lt> Signed-off-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Signed-off-by: Jerry Zhang <jerryzh168@gmail.com> Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai> Signed-off-by: Chengji Yao <chengjiyao@google.com> Signed-off-by: Xu Song <xusong.vip@gmail.com> Signed-off-by: Aaron Pham <contact@aarnphm.xyz> Signed-off-by: Dipika Sikka <dipikasikka1@gmail.com> Signed-off-by: rzou <zou3519@gmail.com> Signed-off-by: Siqi Yan <siqi@meta.com> Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: Nishidha Panpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Signed-off-by: npanpaliya <nishidha.panpaliya@partner.ibm.com> Signed-off-by: Chenyaaang <chenyangli@google.com> Signed-off-by: Alexei V. Ivanov <alexei.ivanov@amd.com> Signed-off-by: ElizaWszola <ewszola@redhat.com> Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com> Signed-off-by: Qiliang Cui <derrhein@gmail.com> Signed-off-by: Aaruni Aggarwal <aaruniagg@gmail.com> Signed-off-by: drisspg <drisspguessous@gmail.com> Signed-off-by: Lifan Shen <lifans@meta.com> Signed-off-by: pramkuma <Pramendra.Kumar@amd.com> Signed-off-by: luka <luka@neuralmagic.com> Signed-off-by: Richard Zou <zou3519@gmail.com> Signed-off-by: Xu Wenqing <xuwq1993@qq.com> Signed-off-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Signed-off-by: yZhen <yZhen@fb.com> Signed-off-by: KsuParkhamchuk <k.parkhamchuk@gmail.com> Signed-off-by: cr7258 <chengzw258@163.com> Signed-off-by: Conroy Cheers <conroy@corncheese.org> Signed-off-by: windsonsea <haifeng.yao@daocloud.io> Signed-off-by: Yinghai Lu <yinghai@thinkingmachines.ai> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> Signed-off-by: liusiqian <liusiqian@tal.com> Signed-off-by: Pavani Majety <pmajety@nvidia.com> Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com> Signed-off-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> Signed-off-by: Xiongfei Wei <isaacwxf23@gmail.com> Signed-off-by: wangli <wangli858794774@gmail.com> Signed-off-by: Anna Pendleton <pendleton@google.com> Signed-off-by: Tsai, Louie <louie.tsai@intel.com> Signed-off-by: Yunqiu Guo <guorachel@meta.com> Signed-off-by: jiang.li <jiang1.li@intel.com> Signed-off-by: youkaichao <youkaichao@gmail.com> Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com> Signed-off-by: py-andy-c <pychen1017@gmail.com> Signed-off-by: niu_he <carlton2tang@gmail.com> Signed-off-by: Junhao Li <junhao@ubicloud.com> Signed-off-by: artetaout <lulala341@gmail.com> Signed-off-by: ximing.wxm <ximing.wxm@antgroup.com> Signed-off-by: Runzhen Wang <wangrunzhen@gmail.com> Signed-off-by: David Xia <david@davidxia.com> Signed-off-by: Bill Nell <bnell@redhat.com> Signed-off-by: Randall Smith <Randall.Smith@amd.com> Signed-off-by: Andy Xie <andy.xning@gmail.com> Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca> Signed-off-by: strutive07 <strutive07@gmail.com> Signed-off-by: 2niuhe <carlton2tang@gmail.com> Signed-off-by: NickLucche <nlucches@redhat.com> Signed-off-by: yewentao256 <zhyanwentao@126.com> Signed-off-by: mobicham <hicham@mobiuslabs.com> Signed-off-by: Luka Govedič <lgovedic@redhat.com> Signed-off-by: qizixi <qizixi@meta.com> Signed-off-by: Zerohertz <ohg3417@gmail.com> Signed-off-by: jiang1.li <jiang1.li@intel.com> Signed-off-by: Boyuan Feng <boyuan@meta.com> Signed-off-by: qingjun <qingjun@minimaxi.com> Signed-off-by: Yida Wu <yidawu@alumni.cmu.edu> Signed-off-by: Saheli Bhattacharjee <saheli@krai.ai> Signed-off-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Signed-off-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Signed-off-by: SzymonOzog <szymon.ozog@gmail.com> Signed-off-by: Shawn Tan <shawntan@ibm.com> Signed-off-by: QscQ <qscqesze@gmail.com> Signed-off-by: Amog Kamsetty <amogkamsetty@gmail.com> Co-authored-by: Raushan Turganbay <raushan.turganbay@alumni.nu.edu.kz> Co-authored-by: Lu Fang <30275821+houseroad@users.noreply.github.com> Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com> Co-authored-by: CYJiang <86391540+googs1025@users.noreply.github.com> Co-authored-by: Simon Mo <simon.mo@hey.com> Co-authored-by: SorenDreano <71752785+SorenDreano@users.noreply.github.com> Co-authored-by: Soren Dreano <soren@numind.ai> Co-authored-by: Reid <61492567+reidliu41@users.noreply.github.com> Co-authored-by: reidliu41 <reid201711@gmail.com> Co-authored-by: Varun Sundar Rabindranath <varunsundar08@gmail.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Yong Hoon Shin <48474650+sarckk@users.noreply.github.com> Co-authored-by: Michael Goin <mgoin64@gmail.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Yikun Jiang <yikun@apache.org> Co-authored-by: Chen Zhang <zhangch99@outlook.com> Co-authored-by: Ekagra Ranjan <3116519+ekagra-ranjan@users.noreply.github.com> Co-authored-by: Chauncey <chaunceyjiang@gmail.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> Co-authored-by: Yan Ru Pei <yanrpei@gmail.com> Co-authored-by: Jiaxin Shan <seedjeffwan@gmail.com> Co-authored-by: Russell Bryant <rbryant@redhat.com> Co-authored-by: Mark McLoughlin <markmc@redhat.com> Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com> Co-authored-by: Li, Jiang <jiang1.li@intel.com> Co-authored-by: Lukas Geiger <lukas.geiger94@gmail.com> Co-authored-by: Vadim Gimpelson <156319763+vadiklyutiy@users.noreply.github.com> Co-authored-by: Calvin Chen <45745657+calvin0327@users.noreply.github.com> Co-authored-by: Kaixi Hou <kaixih@nvidia.com> Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Co-authored-by: 汪志鹏 <wangzhipeng628@gmail.com> Co-authored-by: Siyuan Liu <lsiyuan@google.com> Co-authored-by: Seiji Eicher <58963096+eicherseiji@users.noreply.github.com> Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn> Co-authored-by: wang.yuqi <noooop@126.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk> Co-authored-by: Xu Wenqing <121550081+Xu-Wenqing@users.noreply.github.com> Co-authored-by: Lain <fusiyuan2000@hotmail.com> Co-authored-by: jmswen <jmswen@users.noreply.github.com> Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com> Co-authored-by: Kebe <mail@kebe7jun.com> Co-authored-by: Nick Hill <nhill@redhat.com> Co-authored-by: Yang Wang <elainewy@meta.com> Co-authored-by: Huy Do <huydhn@gmail.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com> Co-authored-by: 22quinn <33176974+22quinn@users.noreply.github.com> Co-authored-by: Guillaume Calmettes <gcalmettes@scaleway.com> Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> Co-authored-by: Chiyue Wei <92623189+dubcyfor3@users.noreply.github.com> Co-authored-by: Chiyue Wei <chiyuew@nvidia.com> Co-authored-by: Povilas Kanapickas <povilas@radix.lt> Co-authored-by: Dipika Sikka <dipikasikka1@gmail.com> Co-authored-by: Luis Vega <vegaluisjose@users.noreply.github.com> Co-authored-by: Luis Vega <2478335+vegaluisjose@users.noreply.github.com> Co-authored-by: Jerry Zhang <jerryzh168@gmail.com> Co-authored-by: Benjamin Chislett <benjamin.chislett@centml.ai> Co-authored-by: Chengji Yao <chengjiyao@google.com> Co-authored-by: Xu Song <xusong.vip@gmail.com> Co-authored-by: Aaron Pham <contact@aarnphm.xyz> Co-authored-by: Jinghui Zhang <jinghuizhang0804@gmail.com> Co-authored-by: jinghui <jinghui@fb.com> Co-authored-by: Richard Zou <zou3519@users.noreply.github.com> Co-authored-by: Siqi Yan <ysq0807@hotmail.com> Co-authored-by: Siqi Yan <siqi@meta.com> Co-authored-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: Yu Guo <82124926+yuguo68@users.noreply.github.com> Co-authored-by: Nishidha <nishidha.panpaliya@partner.ibm.com> Co-authored-by: Md. Shafi Hussain <Md.Shafi.Hussain@ibm.com> Co-authored-by: Adolfo Victoria <adolfokarim@gmail.com> Co-authored-by: Adolfo Victoria <adovi@meta.com> Co-authored-by: Chenyaaang <42742451+Chenyaaang@users.noreply.github.com> Co-authored-by: Alexei-V-Ivanov-AMD <156011006+Alexei-V-Ivanov-AMD@users.noreply.github.com> Co-authored-by: ElizaWszola <ewszola@redhat.com> Co-authored-by: QiliangCui <derrhein@gmail.com> Co-authored-by: Aaruni Aggarwal <47731267+AaruniAggarwal@users.noreply.github.com> Co-authored-by: Driss Guessous <32754868+drisspg@users.noreply.github.com> Co-authored-by: Lifans <draftbks@gmail.com> Co-authored-by: pramenku <7664080+pramenku@users.noreply.github.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Co-authored-by: Akash kaothalkar <61960177+Akashcodes732@users.noreply.github.com> Co-authored-by: Akash Kaothalkar <akash.kaothalkar@ibm.com> Co-authored-by: jennyyyyzhen <47012288+jennyyyyzhen@users.noreply.github.com> Co-authored-by: yZhen <yZhen@fb.com> Co-authored-by: Kseniya Parkhamchuk <43078183+KsuParkhamchuk@users.noreply.github.com> Co-authored-by: Se7en <chengzw258@163.com> Co-authored-by: Conroy Cheers <conroy@corncheese.org> Co-authored-by: Michael Yao <haifeng.yao@daocloud.io> Co-authored-by: Yinghai Lu <yinghai@thinkingmachines.ai> Co-authored-by: Kyle Sayers <kylesayrs@gmail.com> Co-authored-by: liusiqian-tal <141730978+liusiqian-tal@users.noreply.github.com> Co-authored-by: Pavani Majety <pmajety@nvidia.com> Co-authored-by: Ye (Charlotte) Qi <yeq@meta.com> Co-authored-by: Tianyu Guo <guoty9@mail2.sysu.edu.cn> Co-authored-by: XiongfeiWei <isaacwxf23@gmail.com> Co-authored-by: Li Wang <wangli858794774@gmail.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: Anna Pendleton <pendleton@google.com> Co-authored-by: Louie Tsai <louie.tsai@intel.com> Co-authored-by: Li, Jiang <bigpyj64@gmail.com> Co-authored-by: Rachel Guo <35738743+YUNQIUGUO@users.noreply.github.com> Co-authored-by: youkaichao <youkaichao@gmail.com> Co-authored-by: Isotr0py <2037008807@qq.com> Co-authored-by: Gregory Shtrasberg <156009573+gshtras@users.noreply.github.com> Co-authored-by: py-andy-c <37168711+py-andy-c@users.noreply.github.com> Co-authored-by: niu_he <carlton2tang@gmail.com> Co-authored-by: Junhao Li <junhao@ubicloud.com> Co-authored-by: leopardracer <136604165+leopardracer@users.noreply.github.com> Co-authored-by: artetaout <128046886+artetaout@users.noreply.github.com> Co-authored-by: Ximingwang-09 <72070413+Ximingwang-09@users.noreply.github.com> Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com> Co-authored-by: runzhen <wangrunzhen@gmail.com> Co-authored-by: David Xia <david@davidxia.com> Co-authored-by: bnellnm <49004751+bnellnm@users.noreply.github.com> Co-authored-by: rasmith <Randall.Smith@amd.com> Co-authored-by: Ning Xie <andy.xning@gmail.com> Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca> Co-authored-by: wonjun Jang <strutive07@gmail.com> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Co-authored-by: mobicham <37179323+mobicham@users.noreply.github.com> Co-authored-by: Sage Moore <sage@neuralmagic.com> Co-authored-by: kourosh hakhamaneshi <31483498+kouroshHakha@users.noreply.github.com> Co-authored-by: qizixi <22851944+zixi-qi@users.noreply.github.com> Co-authored-by: Hyogeun Oh (오효근) <ohg3417@gmail.com> Co-authored-by: Boyuan Feng <fby.1994@gmail.com> Co-authored-by: qscqesze <qingjun@minimaxi.com> Co-authored-by: Concurrensee <yida.wu@amd.com> Co-authored-by: Saheli Bhattacharjee <47847054+sahelib25@users.noreply.github.com> Co-authored-by: jiahanc <173873397+jiahanc@users.noreply.github.com> Co-authored-by: Konrad Zawora <kzawora@habana.ai> Co-authored-by: maobaolong <baoloongmao@tencent.com> Co-authored-by: Ilya Markov <markovilya197@gmail.com> Co-authored-by: quanliu <33453350+quanliu1991@users.noreply.github.com> Co-authored-by: 刘全 <quan.liu2@dbappsecurity.com.cn> Co-authored-by: Francesco Bertolotti <f14.bertolotti@gmail.com> Co-authored-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Szymon Ożóg <58388001+SzymonOzog@users.noreply.github.com> Co-authored-by: Navanit Dubey <98005188+Navanit-git@users.noreply.github.com> Co-authored-by: Shawn Tan <shawntan@ibm.com> Co-authored-by: qscqesze <qscqesze@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Essential Elements of an Effective PR Description Checklist
Motivation
The
scaled_int8_quantkernel family still relied on scalar loads/stores, leaving significant throughput untapped. Issue #18866 highlighted vectorization as the next major opportunity. Thank @mgoin for the great issue!And thank @ztang2370 for the good start #19062, #19109 by @mgoin for the great incremental compilation document.
What’s in this PR
static_scaled_int8_quant_kernelstatic_scaled_int8_azp_quant_kerneldynamic_scaled_int8_quant_kerneldynamic_scaled_int8_azp_quant_kernelMake a abstraction in
csrc/quantization/vectorization_utils.cuhso that this func can be reused later.Zero-point refactor
Benchmark tooling
benchmarks/kernels/bench_int8_gemm.pyfor side-by-side BF16 vs. INT8 micro-benchmarks.Test
Tested on H100:
Accuracy validation:
lm_eval --model vllm --model_args pretrained=RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8,max_model_len=32768 --trust_remote_code --tasks gsm8k --num_fewshot 5 --batch_size autoEnd to end throughput:
vllm bench throughput --model RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8 --load-format dummy --input-len 1000 --output-len 100 --max-model-len 32768Kernel flops test with
bench_int8_gemm.pyBefore:
After:
Unit test: