vllm 0.8.3

v0.8.3

on Python PyPI

Highlights

This release features 260 commits, 109 contributors, 38 new contributors.

• We are excited to announce Day 0 Support for Llama 4 Scout and Maverick (#16104). Please see our blog for detailed user guide.
  • Please note that Llama4 is only supported in V1 engine only for now.
• V1 engine now supports native sliding window attention (#14097) with the hybrid memory allocator.

Cluster Scale Serving

• Single node data parallel with API server support (#13923)
• Multi-node offline DP+EP example (#15484)
• Expert parallelism enhancements
  • CUTLASS grouped gemm fp8 MoE kernel (#13972)
  • Fused experts refactor (#15914)
  • Fp8 Channelwise Dynamic Per Token GroupedGEMM (#15587)
  • Adding support for fp8 gemm layer input in fp8 (#14578)
  • Add option to use DeepGemm contiguous grouped gemm kernel for fused MoE operations. (#13932)
• Support XpYd disaggregated prefill with MooncakeStore (#12957)

Model Supports

• Llama 4 (#16104), Aya Vision (#15441), MiniMaxText01(#13454), Skywork-R1V (#15397), jina-reranker-v2 (#15876)
• Add Reasoning Parser for Granite Models (#14202)
• Add Phi-4-mini function calling support (#14886)

V1 Engine

• Collective RPC (#15444)
• Faster top-k only implementation (#15478)
• BitsAndBytes support (#15611)
• Speculative Decoding: metrics (#15151), Eagle Proposer (#15729), n-gram interface update (#15750), EAGLE Architecture with Proper RMS Norms (#14990)

Features

API

• Support Enum for xgrammar based structured output in V1. (#15594, #15757)
• A new tags parameter for wake_up (#15500)
• V1 LoRA support CPU offload (#15843)
• Prefix caching support: FIPS enabled machines with MD5 hashing (#15299), SHA256 as alternative hashing algorithm (#15297)
• Addition of http service metrics (#15657)

Performance

• LoRA Scheduler optimization bridging V1 and V0 performance (#15422).

Hardwares

• AMD:
  • Add custom allreduce support for ROCM (#14125)
  • Quark quantization documentation (#15861)
  • AITER integration: int8 scaled gemm kernel (#15433), fused moe (#14967)
  • Paged attention for V1 (#15720)
• CPU:
  • CPU MLA (#14744)
• TPU
  • Improve Memory Usage Estimation (#15671)
  • Optimize the all-reduce performance (#15903)
  • Support sliding window and logit soft capping in the paged attention kernel. (#15732)
  • TPU-optimized top-p implementation (avoids scattering). (#15736)

Doc, Build, Ecosystem

• V1 user guide update: fp8 kv cache support (#15585), multi-modality (#15460)
• Recommend developing with Python 3.12 in developer guide (#15811)
• Clean up: move dockerfiles into their own directory (#14549)
• Add minimum version for huggingface_hub to enable Xet downloads (#15873)
• TPU CI: Add basic perf regression test (#15414)

What's Changed

• Fix CUDA kernel index data type in vllm/csrc/quantization/gptq_marlin/awq_marlin_repack.cu +10 by @houseroad in #15160
• [Hardware][TPU][Bugfix] Fix v1 mp profiler by @lsy323 in #15409
• [Kernel][CPU] CPU MLA by @gau-nernst in #14744
• Dockerfile.ppc64le changes to move to UBI by @Shafi-Hussain in #15402
• [Misc] Clean up MiniCPM-V/O code by @DarkLight1337 in #15337
• [Misc] Remove redundant num_embeds by @DarkLight1337 in #15443
• [Doc] Update V1 user guide for multi-modality by @DarkLight1337 in #15460
• [Kernel] Fix conflicting macro names for gguf kernels by @SzymonOzog in #15456
• [bugfix] fix inductor cache on max_position_embeddings by @youkaichao in #15436
• [CI/Build] Add tests for the V1 tpu_model_runner. by @yarongmu-google in #14843
• [Bugfix] Support triton==3.3.0+git95326d9f for RTX 5090 (Unsloth + vLLM compatibility) by @oteroantoniogom in #15471
• [bugfix] add supports_v1 platform interface by @joerunde in #15417
• Add workaround for shared field_names in pydantic model class by @maxdebayser in #13925
• [TPU][V1] Fix Sampler recompilation by @NickLucche in #15309
• [V1][Minor] Use SchedulerInterface type for engine scheduler field by @njhill in #15499
• [V1] Support long_prefill_token_threshold in v1 scheduler by @houseroad in #15419
• [core] add bucket padding to tpu_model_runner by @Chenyaaang in #14995
• [Core] LoRA: V1 Scheduler optimization by @varun-sundar-rabindranath in #15422
• [CI/Build] LoRA: Delete long context tests by @varun-sundar-rabindranath in #15503
• Transformers backend already supports V1 by @hmellor in #15463
• [Model] Support multi-image for Molmo by @DarkLight1337 in #15438
• [Misc] Warn about v0 in benchmark_paged_attn.py by @tlrmchlsmth in #15495
• [BugFix] Fix nightly MLA failure (FA2 + MLA chunked prefill, i.e. V1, producing bad results) by @LucasWilkinson in #15492
• [misc] LoRA - Skip LoRA kernels when not required by @varun-sundar-rabindranath in #15152
• Fix raw_request extraction in load_aware_call decorator by @daniel-salib in #15382
• [Feature] Enhance EAGLE Architecture with Proper RMS Norms by @luyuzhe111 in #14990
• [FEAT][ROCm] Integrate Fused MoE Kernels from AITER by @vllmellm in #14967
• [Misc] Enhance warning information to user-defined chat template by @wwl2755 in #15408
• [Misc] improve example script output by @reidliu41 in #15528
• Separate base model from TransformersModel by @hmellor in #15467
• Apply torchfix by @cyyever in #15532
• Improve validation of TP in Transformers backend by @hmellor in #15540
• [Model] Add Reasoning Parser for Granite Models by @alex-jw-brooks in #14202
• multi-node offline DP+EP example by @youkaichao in #15484
• Fix weight loading for some models in Transformers backend by @hmellor in #15544
• [Refactor] Remove passthrough backend when generate grammar by @aarnphm in #15317
• [V1][Sampler] Faster top-k only implementation by @njhill in #15478
• Support SHA256 as hash function in prefix caching by @dr75 in #15297
• Applying some fixes for K8s agents in CI by @Alexei-V-Ivanov-AMD in #15493
• [V1] TPU - Revert to exponential padding by default by @alexm-redhat in #15565
• [V1] TPU CI - Fix test_compilation.py by @alexm-redhat in #15570
• Use Cache Hinting for fused_moe kernel by @wrmedford in #15511
• [TPU] support disabling xla compilation cache by @yaochengji in #15567
• Support FIPS enabled machines with MD5 hashing by @MattTheCuber in #15299
• [Kernel] CUTLASS grouped gemm fp8 MoE kernel by @ElizaWszola in #13972
• Add automatic tpu label to mergify.yml by @mgoin in #15560
• add platform check back by @Chenyaaang in #15578
• [misc] LoRA: Remove unused long context test data by @varun-sundar-rabindranath in #15558
• [Doc] Update V1 user guide for fp8 kv cache support by @wayzeng in #15585
• [moe][quant] add weight name case for offset by @MengqingCao in #15515
• [V1] Refactor num_computed_tokens logic by @comaniac in #15307
• Allow torchao quantization in SiglipMLP by @jerryzh168 in #15575
• [ROCm] Env variable to trigger custom PA by @gshtras in #15557
• [TPU] [V1] fix cases when max_num_reqs is set smaller than MIN_NUM_SEQS by @yaochengji in #15583
• [Misc] Restrict ray version dependency and update PP feature warning in V1 by @ruisearch42 in #15556
• [TPU] Avoid Triton Import by @robertgshaw2-redhat in #15589
• [Misc] Consolidate LRUCache implementations by @Avabowler in #15481
• [Quantization] Fp8 Channelwise Dynamic Per Token GroupedGEMM by @robertgshaw2-redhat in #15587
• [Misc] Clean up scatter_patch_features by @DarkLight1337 in #15559
• [Misc] Use model_redirect to redirect the model name to a local folder. by @noooop in #14116
• Fix incorrect filenames in vllm_compile_cache.py by @zou3519 in #15494
• [Doc] update --system for transformers installation in docker doc by @reidliu41 in #15616
• [Model] MiniCPM-V/O supports V1 by @DarkLight1337 in #15487
• [Bugfix] Fix use_cascade_attention handling for Alibi-based models on vllm/v1 by @h-sugi in #15211
• [Doc] Link to onboarding tasks by @DarkLight1337 in #15629
• [Misc] Replace is_encoder_decoder_inputs with split_enc_dec_inputs by @DarkLight1337 in #15620
• [Feature] Add middleware to log API Server responses by @terrytangyuan in #15593
• [Misc] Avoid direct access of global mm_registry in compute_encoder_budget by @DarkLight1337 in #15621
• [Doc] Use absolute placement for Ask AI button by @hmellor in #15628
• [Bugfix][TPU][V1] Fix recompilation by @NickLucche in #15553
• Correct PowerPC to modern IBM Power by @clnperez in #15635
• [CI] Update rules for applying tpu label. by @russellb in #15634
• [V1] AsyncLLM data parallel by @njhill in #13923
• [TPU] Lazy Import by @robertgshaw2-redhat in #15656
• [Quantization][V1] BitsAndBytes support V1 by @jeejeelee in #15611
• [Bugfix] Fix failure to launch in Tensor Parallel TP mode on macOS. by @kebe7jun in #14948
• [Doc] Fix dead links in Job Board by @wwl2755 in #15637
• [CI][TPU] Temporarily Disable Quant Test on TPU by @robertgshaw2-redhat in #15649
• Revert "Use Cache Hinting for fused_moe kernel (#15511)" by @wrmedford in #15645
• [Misc]add coding benchmark for speculative decoding by @CXIAAAAA in #15303
• [Quantization][FP8] Adding support for fp8 gemm layer input in fp8 by @gshtras in #14578
• Refactor error handling for multiple exceptions in preprocessing by @JasonZhu1313 in #15650
• [Bugfix] Fix mm_hashes forgetting to be passed by @DarkLight1337 in #15668
• [V1] Remove legacy input registry by @DarkLight1337 in #15673
• [TPU][CI] Fix TPUModelRunner Test by @robertgshaw2-redhat in #15667
• [Refactor][Frontend] Keep all logic about reasoning into one class by @gaocegege in #14428
• [CPU][CI] Improve CPU Dockerfile by @bigPYJ1151 in #15690
• [Bugfix] Fix 'InductorAdaptor object has no attribute 'cache_dir' by @jeejeelee in #15674
• [Misc] Fix test_sleep to use query parameters by @lizzzcai in #14373
• [Bugfix][Frontend] Eliminate regex based check in reasoning full generator by @gaocegege in #14821
• [Frontend] update priority for --api-key and VLLM_API_KEY by @reidliu41 in #15588
• [Docs] Add "Generation quality changed" section to troubleshooting by @hmellor in #15701
• [Model] Adding torch compile annotations to chatglm by @jeejeelee in #15624
• [Bugfix][v1] xgrammar structured output supports Enum. by @chaunceyjiang in #15594
• [Bugfix] embed_is_patch for Idefics3 by @DarkLight1337 in #15696
• [V1] Support disable_any_whtespace for guidance backend by @russellb in #15584
• [doc] add missing imports by @reidliu41 in #15699
• [Bugfix] Fix regex compile display format by @kebe7jun in #15368
• Fix cpu offload testing for gptq/awq/ct by @mgoin in #15648
• [Minor] Remove TGI launching script by @WoosukKwon in #15646
• [Misc] Remove unused utils and clean up imports by @DarkLight1337 in #15708
• [Misc] Remove stale func in KVTransferConfig by @ShangmingCai in #14746
• [TPU] [Perf] Improve Memory Usage Estimation by @robertgshaw2-redhat in #15671
• [Bugfix] [torch.compile] Add Dynamo metrics context during compilation by @ProExpertProg in #15639
• [V1] TPU - Fix the chunked prompt bug by @alexm-redhat in #15713
• [Misc] cli auto show default value by @reidliu41 in #15582
• implement prometheus fast-api-instrumentor for http service metrics by @daniel-salib in #15657
• [Docs][V1] Optimize diagrams in prefix caching design by @simpx in #15716
• [ROCm][AMD][Build] Update AMD supported arch list by @gshtras in #15632
• [Model] Support Skywork-R1V by @pengyuange in #15397
• [Docs] Document v0 engine support in reasoning outputs by @gaocegege in #15739
• [Misc][V1] Misc code streamlining by @njhill in #15723
• [Bugfix] LoRA V1: add and fix entrypoints tests by @varun-sundar-rabindranath in #15715
• [CI] Speed up V1 structured output tests by @russellb in #15718
• Use numba 0.61 for python 3.10+ to support numpy>=2 by @cyyever in #15692
• [Bugfix] set VLLM_WORKER_MULTIPROC_METHOD=spawn for vllm.entrypoionts.openai.api_server by @jinzhen-lin in #15700
• [TPU][V1][Bugfix] Fix w8a8 recompiilation with GSM8K by @NickLucche in #15714
• [Kernel][TPU][ragged-paged-attn] vLLM code change for PR#8896 by @yarongmu-google in #15659
• [doc] update doc by @reidliu41 in #15740
• [FEAT] [ROCm] Add AITER int8 scaled gemm kernel by @tjtanaa in #15433
• [V1] [Feature] Collective RPC by @wwl2755 in #15444
• [Feature][Disaggregated] Support XpYd disaggregated prefill with MooncakeStore by @ShangmingCai in #12957
• [V1] Support interleaved modality items by @ywang96 in #15605
• [V1][Minor] Simplify rejection sampler's parse_output by @WoosukKwon in #15741
• [Bugfix] Fix Mllama interleaved images input support by @Isotr0py in #15564
• [CI] xgrammar structured output supports Enum. by @chaunceyjiang in #15757
• [Bugfix] Fix Mistral guided generation using xgrammar by @juliendenize in #15704
• [doc] update conda to usage link in installation by @reidliu41 in #15761
• fix test_phi3v by @pansicheng in #15321
• [V1] Override mm_counts for dummy data creation by @DarkLight1337 in #15703
• fix: lint fix a ruff checkout syntax error by @yihong0618 in #15767
• [Bugfix] Added embed_is_patch mask for fuyu model by @kylehh in #15731
• fix: Comments to English for better dev experience by @yihong0618 in #15768
• [V1][Scheduler] Avoid calling _try_schedule_encoder_inputs for every request by @WoosukKwon in #15778
• [Misc] update the comments by @lcy4869 in #15780
• [Benchmark] Update Vision Arena Dataset and HuggingFaceDataset Setup by @JenZhao in #15748
• [Feature][ROCm]Enable fusion pass for torch.compile on ROCm by @charlifu in #15050
• Recommend developing with Python 3.12 in developer guide by @hmellor in #15811
• fix: better install requirement for install in setup.py by @yihong0618 in #15796
• [V1] Fully Transparent Implementation of CPU Offloading by @youkaichao in #15354
• [Model] Update support for NemotronNAS models by @Naveassaf in #15008
• [Bugfix] Fix Crashing When Loading Modules With Batchnorm Stats by @alex-jw-brooks in #15813
• [Bugfix] Fix missing return value in load_weights method of adapters.py by @noc-turne in #15542
• Upgrade transformers to v4.50.3 by @hmellor in #13905
• [Bugfix] Check dimensions of multimodal embeddings in V1 by @DarkLight1337 in #15816
• [V1][Spec Decode] Remove deprecated spec decode config params by @ShangmingCai in #15466
• fix: change GB to GiB in logging close #14979 by @yihong0618 in #15807
• [V1] TPU CI - Add basic perf regression test by @alexm-redhat in #15414
• Fix Transformers backend compatibility check by @hmellor in #15290
• [V1][Core] Remove unused speculative config from scheduler by @markmc in #15818
• Move dockerfiles into their own directory by @hmellor in #14549
• [Distributed] Add custom allreduce support for ROCM by @ilmarkov in #14125
• Rename fallback model and refactor supported models section by @hmellor in #15829
• [Frontend] Add Phi-4-mini function calling support by @kinfey in #14886
• [Bugfix][Model] fix mllama multi-image by @yma11 in #14883
• [Bugfix] Fix extra comma by @haochengxia in #15851
• [Bugfix]: Fix is_embedding_layer condition in VocabParallelEmbedding by @alexwl in #15824
• [V1] TPU - Fix fused MOE by @alexm-redhat in #15834
• [sleep mode] clear pytorch cache after sleep by @lionelvillard in #15248
• [ROCm] Use device name in the warning by @gshtras in #15838
• [V1] Implement sliding window attention in kv_cache_manager by @heheda12345 in #14097
• fix: can not use uv run collect_env close #13888 by @yihong0618 in #15792
• [Feature] specify model in config.yaml by @wayzeng in #15798
• [Misc] Enable V1 LoRA by default by @varun-sundar-rabindranath in #15320
• [Misc] Fix speculative config repr string by @ShangmingCai in #15860
• [Docs] Fix small error in link text by @hmellor in #15868
• [Bugfix] Fix no video/image profiling edge case for MultiModalDataParser by @Isotr0py in #15828
• [Misc] Use envs.VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE by @ruisearch42 in #15831
• setup correct nvcc version with CUDA_HOME by @chenyang78 in #15725
• [Model] Support Mistral3 in the HF Transformers format by @mgoin in #15505
• [Misc] remove unused script by @reidliu41 in #15746
• Remove format.sh as it's been unsupported >70 days by @hmellor in #15884
• [New Model]: jinaai/jina-reranker-v2-base-multilingual by @noooop in #15876
• [Doc] Quark quantization documentation by @cha557 in #15861
• Reinstate format.sh and make pre-commit installation simpler by @hmellor in #15890
• [Misc] Allow using OpenCV as video IO fallback by @Isotr0py in #15055
• [ROCm][Build][Bugfix] Bring the base dockerfile in sync with the ROCm fork by @gshtras in #15820
• Add option to use DeepGemm contiguous grouped gemm kernel for fused MoE operations. by @bnellnm in #13932
• [CI/Build] Clean up LoRA tests by @jeejeelee in #15867
• [Model] Aya Vision by @JenZhao in #15441
• [Model] Add module name prefixes to gemma3 by @cloud11665 in #15889
• [CI] Disable flaky structure decoding test temporarily. by @ywang96 in #15892
• [V1][Metrics] Initial speculative decoding metrics by @markmc in #15151
• [V1][Spec Decode] Implement Eagle Proposer [1/N] by @WoosukKwon in #15729
• [Docs] update usage stats language by @simon-mo in #15898
• [BugFix] make sure socket close by @yihong0618 in #15875
• [Model][MiniMaxText01] Support MiniMaxText01 model inference by @ZZBoom in #13454
• [Docs] Add Ollama meetup slides by @simon-mo in #15905
• [Docs] Add Intel as Sponsor by @simon-mo in #15913
• Fix input triton kernel for eagle by @ekagra-ranjan in #15909
• [V1] Fix: make sure k_index is int64 for apply_top_k_only by @b8zhong in #15907
• [Bugfix] Fix imports for MoE on CPU by @gau-nernst in #15841
• [V1][Minor] Enhance SpecDecoding Metrics Log in V1 by @WoosukKwon in #15902
• [Doc] Update rocm.inc.md by @chun37 in #15917
• [V1][Bugfix] Fix typo in MoE TPU checking by @ywang96 in #15927
• [Benchmark]Fix error message by @Potabk in #15866
• [Misc] Replace print with logger by @chaunceyjiang in #15923
• [CI/Build] Further clean up LoRA tests by @jeejeelee in #15920
• [Bugfix] Fix cache block size calculation for CPU MLA by @gau-nernst in #15848
• [Build/CI] Update lm-eval to 0.4.8 by @cthi in #15912
• [Kernel] Add more dtype support for GGUF dequantization by @LukasBluebaum in #15879
• [core] Add tags parameter to wake_up() by @erictang000 in #15500
• [V1] Fix json_object support with xgrammar by @russellb in #15488
• Add minimum version for huggingface_hub to enable Xet downloads by @hmellor in #15873
• [Bugfix][Benchmarks] Ensure async_request_deepspeed_mii uses the OpenAI choices key by @b8zhong in #15926
• [CI] Remove duplicate entrypoints-test by @yankay in #15940
• [Bugfix] Fix the issue where the model name is empty string, causing no response with the model name. by @chaunceyjiang in #15938
• [Metrics] Hide deprecated metrics by @markmc in #15458
• [Frontend] Implement Tool Calling with tool_choice='required' by @meffmadd in #13483
• [CPU][Bugfix] Using custom allreduce for CPU backend by @bigPYJ1151 in #15934
• [Model] use AutoWeightsLoader in model load_weights by @lengrongfu in #15770
• [Misc] V1 LoRA support CPU offload by @jeejeelee in #15843
• Restricted cmake to be less than version 4 as 4.x breaks the build of… by @npanpaliya in #15859
• [misc] instruct pytorch to use nvml-based cuda check by @youkaichao in #15951
• [V1] Support Mistral3 in V1 by @mgoin in #15950
• Fix huggingface-cli[hf-xet] -> huggingface-cli[hf_xet] by @hmellor in #15969
• [V1][TPU] TPU-optimized top-p implementation (avoids scattering). by @hyeygit in #15736
• [TPU] optimize the all-reduce performance by @yaochengji in #15903
• [V1][TPU] Do not compile sampling more than needed by @NickLucche in #15883
• [ROCM][KERNEL] Paged attention for V1 by @maleksan85 in #15720
• fix: better error message for get_config close #13889 by @yihong0618 in #15943
• [bugfix] add seed in torchrun_example.py by @youkaichao in #15980
• [ROCM][V0] PA kennel selection when no sliding window provided by @maleksan85 in #15982
• [Benchmark] Add AIMO Dataset to Benchmark by @StevenShi-23 in #15955
• [misc] improve error message for "Failed to infer device type" by @youkaichao in #15994
• [Bugfix][V1] Fix bug from putting llm_engine.model_executor in a background process by @wwl2755 in #15367
• [doc] update contribution link by @reidliu41 in #15922
• fix: tiny fix make format.sh excutable by @yihong0618 in #16015
• [SupportsQuant] Bert, Blip, Blip2, Bloom by @kylesayrs in #15573
• [SupportsQuant] Chameleon, Chatglm, Commandr by @kylesayrs in #15952
• [Neuron][kernel] Fuse kv cache into a single tensor by @liangfu in #15911
• [Minor] Fused experts refactor by @bnellnm in #15914
• [Misc][Performance] Advance tpu.txt to the most recent nightly torch … by @yarongmu-google in #16024
• Re-enable the AMD Testing for the passing tests. by @Alexei-V-Ivanov-AMD in #15586
• [TPU] Support sliding window and logit soft capping in the paged attention kernel for TPU. by @vanbasten23 in #15732
• [TPU] Switch Test to Non-Sliding Window by @robertgshaw2-redhat in #15981
• [Bugfix] Fix function names in test_block_fp8.py by @bnellnm in #16033
• [ROCm] Tweak the benchmark script to run on ROCm by @huydhn in #14252
• [Misc] improve gguf check by @reidliu41 in #15974
• [TPU][V1] Remove ragged attention kernel parameter hard coding by @yaochengji in #16041
• doc: add info for macos clang errors by @yihong0618 in #16049
• [V1][Spec Decode] Avoid logging useless nan metrics by @markmc in #16023
• [Model] use AutoWeightsLoader for baichuan, gpt-neox, mpt by @jonghyunchoe in #15939
• [Hardware][Gaudi][BugFix] fix arguments of hpu fused moe by @zhenwei-intel in #15945
• [Bugfix][kernels] Fix half2float conversion in gguf kernels by @Isotr0py in #15995
• [Benchmark][Doc] Update throughput benchmark and README by @StevenShi-23 in #15998
• [CPU] Change default block_size for CPU backend by @bigPYJ1151 in #16002
• [Distributed] [ROCM] Fix custom allreduce enable checks by @ilmarkov in #16010
• [ROCm][Bugfix] Use platform specific FP8 dtype by @gshtras in #15717
• [ROCm][Bugfix] Bring back fallback to eager mode removed in #14917, but for ROCm only by @gshtras in #15413
• [Bugfix] Fix default behavior/fallback for pp in v1 by @mgoin in #16057
• [CI] Reorganize .buildkite directory by @khluu in #16001
• [V1] DP scale-out (1/N): Use zmq ROUTER/DEALER sockets for input queue by @njhill in #15906
• [V1] Scatter and gather placeholders in the model runner by @DarkLight1337 in #15712
• Revert "[V1] Scatter and gather placeholders in the model runner" by @ywang96 in #16075
• [Kernel][Bugfix] Re-fuse triton moe weight application by @bnellnm in #16071
• [Bugfix][TPU] Fix V1 TPU worker for sliding window by @mgoin in #16059
• [V1][Spec Decode] Update N-gram Proposer Interface by @WoosukKwon in #15750
• [Model] Support Llama4 in vLLM by @houseroad in #16104

New Contributors

• @Shafi-Hussain made their first contribution in #15402
• @oteroantoniogom made their first contribution in #15471
• @cyyever made their first contribution in #15532
• @dr75 made their first contribution in #15297
• @wrmedford made their first contribution in #15511
• @MattTheCuber made their first contribution in #15299
• @jerryzh168 made their first contribution in #15575
• @Avabowler made their first contribution in #15481
• @zou3519 made their first contribution in #15494
• @h-sugi made their first contribution in #15211
• @clnperez made their first contribution in #15635
• @kebe7jun made their first contribution in #14948
• @CXIAAAAA made their first contribution in #15303
• @lizzzcai made their first contribution in #14373
• @simpx made their first contribution in #15716
• @pengyuange made their first contribution in #15397
• @pansicheng made their first contribution in #15321
• @lcy4869 made their first contribution in #15780
• @Naveassaf made their first contribution in #15008
• @noc-turne made their first contribution in #15542
• @ilmarkov made their first contribution in #14125
• @kinfey made their first contribution in #14886
• @haochengxia made their first contribution in #15851
• @alexwl made their first contribution in #15824
• @lionelvillard made their first contribution in #15248
• @cha557 made their first contribution in #15861
• @cloud11665 made their first contribution in #15889
• @ZZBoom made their first contribution in #13454
• @ekagra-ranjan made their first contribution in #15909
• @chun37 made their first contribution in #15917
• @cthi made their first contribution in #15912
• @LukasBluebaum made their first contribution in #15879
• @erictang000 made their first contribution in #15500
• @yankay made their first contribution in #15940
• @meffmadd made their first contribution in #13483
• @lengrongfu made their first contribution in #15770
• @StevenShi-23 made their first contribution in #15955

Full Changelog: v0.8.2...v0.8.3