[Kernel] [Quantization] Add MXFP4 and bias support for marlin kernel#22428

Merged

simon-mo merged 1081 commits intovllm-project:mainfrom

jinzhen-lin:marlin-mxfp4-bias

Aug 14, 2025

This pull request is big! We're only showing the most recent 250 commits

Commits on Aug 9, 2025

[Bugfix] fix when skip tokenizer init (#21922)

lengrongfu
authored and
jinzhen-lin
committed
security policy: take 1 (#21119)

authored and
jinzhen-lin
committed
[Bugfix] [Performance] DeepEPHighThroughput + DeepSeek : Quant before Dispatch (#21837)

authored and
jinzhen-lin
committed
Enable headless models for pooling in the Transformers backend (#21767)

hmellor
authored and
jinzhen-lin
committed
[Misc] Minor enhancement of benchmark_moe (#22068)

jeejeelee
authored and
jinzhen-lin
committed
Fix pre-commit failure for SECURTIY.md (#22102)

mgoin
authored and
jinzhen-lin
committed
[compile][startup] Disable C++ compilation of symbolic shapes (#20836)

anijain2305
authored and
jinzhen-lin
committed
Introduce RayPPCommunicator for ray-based PP (#21660)

ruisearch42
authored and
jinzhen-lin
committed
Add lora test for tp>1 case for TPU. (#21970)

vanbasten23
authored and
jinzhen-lin
committed
[BugFix] Harden distributed DP startup (#21538)

njhill
authored and
jinzhen-lin
committed
[CI] Initial tests for SM100 Blackwell runner (#21877)

mgoin
authored and
jinzhen-lin
committed
[Perf] Optimize `reshape_and_cache_flash` CUDA Kernel (#22036)

yewentao256
authored and
jinzhen-lin
committed
feat: Add Support GPTQ Quantization MOE on ROCM vllm serve (#21733)

JartX
authored and
jinzhen-lin
committed
[V1][CUDA] Full cudagraph support for FlashInfer (#21367)

fhl2000
authored and
jinzhen-lin
committed
[Model] Qwen2.5 VL SiLU-and-Mul (#22066)

authored and
jinzhen-lin
committed
[Misc] `VLLM_TARGET_DEVICE.lower()` (#22101)

NickLucche
authored and
jinzhen-lin
committed
[Misc] DeepGemmExperts : Avoid JIT generation in the hot-path (#21955)

authored and
jinzhen-lin
committed
[Speculators][Speculative Decoding] Add Qwen Eagle3 Support (#21835)

dsikka
authored and
jinzhen-lin
committed
[BugFix] Improve internal DP load balancing (#21617)

njhill
authored and
jinzhen-lin
committed
[Test] Add Unit Test for Batched DeepGEMM (#21559)

yewentao256
authored and
jinzhen-lin
committed
[Attention][DBO] Add support for "splitting" the CommonAttentionMetadata (#21153)

SageMoore
authored and
jinzhen-lin
committed
[FEAT][ROCm] Enable running Flash Attention as ViT attn backend for Qwen-VL models on ROCm platform. (#22069)

authored and
jinzhen-lin
committed
[Misc] Getting and passing ray runtime_env to workers (#22040)

ruisearch42
authored and
jinzhen-lin
committed
Fix test_kv_sharing_fast_prefill flakiness (#22038)

sarckk
authored and
jinzhen-lin
committed
[Bugfix] Mamba2 remove bugged initial state condition in chunk scan (#22034)

cyang49
authored and
jinzhen-lin
committed
docs: remove deprecated disable-log-requests flag (#22113)

Roger Wang
authored and
jinzhen-lin
committed
[PERF] Use faster way of decode in tokenizer: avoid useless list-to-list conversion (#20000)

vadiklyutiy
authored and
jinzhen-lin
committed
for glm-4.1V update (#22000)

authored and
jinzhen-lin
committed
[Model] Mamba2 preallocate SSM output tensor to avoid d2d copy overhead (#21075)

cyang49
authored and
jinzhen-lin
committed
[Frontend] Improve error message for too many mm items (#22114)

DarkLight1337
authored and
jinzhen-lin
committed
[V1] [Hybrid] Validate compatibility of attention backend batch reordering at init time (#21557)

tdoublep
authored and
jinzhen-lin
committed
[xpu]support moe models on XPU platform (#21643)

yma11
authored and
jinzhen-lin
committed
Revert "[compile][startup] Disable C++ compilation of symbolic shapes" (#22122)

xiszishu
authored and
jinzhen-lin
committed
[Misc] Bump ray to 2.48.0 (#22123)

ruisearch42
authored and
jinzhen-lin
committed
[Fix] Fix llama4 modelopt weight loading error (#22107)

authored and
jinzhen-lin
committed
[Misc] Add tensor schema test coverage for multimodal models (#21754)

Isotr0py
authored and
jinzhen-lin
committed
[Benchmark] Support ready check timeout in `vllm bench serve` (#21696)

authored and
jinzhen-lin
committed
Support CUTLASS NVFP4 (w4a4) for Blackwell Geforce GPUs (SM120) (#21309)

LopezCastroRoberto
authored and
jinzhen-lin
committed
[Misc] update doc comment for send (#22026)

andyxning
authored and
jinzhen-lin
committed
[executor] feat: add supports_pp attr to executors (#21786)

eric-haibin-lin
authored and
jinzhen-lin
committed
[V1] [P/D] Refactor KV Connector Path (#21980)

authored and
jinzhen-lin
committed
[Responses API] Disable response store by default (#22137)

WoosukKwon
authored and
jinzhen-lin
committed
[CI/Build][Bugfix] Fix Qwen2.5 tests in CPU CI via fallback silu_and_mul to torch native implementation (#22145)

bigPYJ1151
authored and
jinzhen-lin
committed
Add chat doc in quick start (#21213)

authored and
jinzhen-lin
committed
fuse fp32 for GLM-4.5 e_score_correction_bias (#22143)

zRzRzRzRzRzRzR
authored and
jinzhen-lin
committed
[Bugfix] Fix failing multimodal standard test (#22153)

Isotr0py
authored and
jinzhen-lin
committed
Use `aiohttp` connection pool for benchmarking (#21981)

eicherseiji
authored and
jinzhen-lin
committed
[fix] fix correct assertion syntax error in attention utils. (#22154)

skyloevil
authored and
jinzhen-lin
committed
[RLHF] Fix torch.dtype not serializable in example (#22158)

22quinn
authored and
jinzhen-lin
committed
[PD] add test for chat completions endpoint (#21925)

Abirdcfly
authored and
jinzhen-lin
committed
remove duplicate code within cleanup_dist_env_and_memory (#22147)

andyxning
authored and
jinzhen-lin
committed
Add tree attention backend for v1 (part 1) (#20401)

TheEpicDolphin
authored and
jinzhen-lin
committed
[refactor] improve ConstantList exception specificity (#22156)

skyloevil
authored and
jinzhen-lin
committed
Remove index_put from MM embeddings merging (#22105)

authored and
jinzhen-lin
committed
[CI Bugfix] Fix wNa16 kernel not found for test_shared_storage_connector_hashes (#22163)

tlrmchlsmth
authored and
jinzhen-lin
committed
[Misc] Modify the organization of GLM series (#22171)

jeejeelee
authored and
jinzhen-lin
committed
[feat] move WEIGHT_SCALE_SUPPORTED into raise block to accelerate RLHF weight loading (#21164)

weixiao-huang
authored and
jinzhen-lin
committed
[Bugfix] Fix failing GGUF models test (#22174)

Isotr0py
authored and
jinzhen-lin
committed
[Sampler] Support returning all logprobs or logits (#21792)

22quinn
authored and
jinzhen-lin
committed
[Doc] Update pooling model docs (#22186)

DarkLight1337
authored and
jinzhen-lin
committed
Fix Arcee model weight loading: Add custom load_weights (#21725)

alyosha-swamy
authored and
jinzhen-lin
committed
[Responses API] Ignore `store=True` and process the request by default (#22185)

WoosukKwon
authored and
jinzhen-lin
committed
[Bug] Update auto_tune.sh to separate benchmarking and profiling. (#21629)

ericehanley
authored and
jinzhen-lin
committed
[Bugfix][V1][P/D]Fix the uneven polling issue in the toy proxy for P2pNcclConnector (#21819)

Abatom
authored and
jinzhen-lin
committed
[NVIDIA] Auto detect modelopt quant and fix DSR1-FP4 weight loading (#22073)

nvpohanh
authored and
jinzhen-lin
committed
[Bugfix] V1 Fix the cursor leakage issue during request scheduling. (#21173)

CLFutureX
authored and
jinzhen-lin
committed
Revert "[Bugfix] V1 Fix the cursor leakage issue during request scheduling." (#22223)

WoosukKwon
authored and
jinzhen-lin
committed
[V1] reduce block size for tree attention correctness test to fix 'ou… (#22207)

TheEpicDolphin
authored and
jinzhen-lin
committed
[V0 deprecation][P/D] Deprecate v0 `KVConnectorBase` code (1/2) (#21785)

lk-chen
authored and
jinzhen-lin
committed
[FEAT] Refactor ROPE into module (#22192)

tjtanaa
authored and
jinzhen-lin
committed
[ROCm][Bugfix] Compilation passes fix (#22202)

gshtras
authored and
jinzhen-lin
committed
self.gate dtype update for GLM-4.5 (#22203)

zRzRzRzRzRzRzR
authored and
jinzhen-lin
committed
[Log] DeepGEMM Update Log for Unaligned Problem Size (#22208)

yewentao256
authored and
jinzhen-lin
committed
fix: kimi_k2 return empty tool call list (#22149)

tlipoca9
authored and
jinzhen-lin
committed
[Misc] Remove pass_config from CompilationConfig dump_json excluded (#21911)

elvischenv
authored and
jinzhen-lin
committed
[Doc] add backend to doc string of initialize_model_parallel (#22142)

andyxning
authored and
jinzhen-lin
committed
[Misc] log more detailed message for ensure_model_parallel_initialized (#22144)

andyxning
authored and
jinzhen-lin
committed
Optimize configuration access with LRU cache in custom ops (#22204)

skyloevil
authored and
jinzhen-lin
committed
[Bugfix] Misaligned params in TreeAttentionImpl (#22226)

DarkLight1337
authored and
jinzhen-lin
committed
[UX] Fail if an invalid attention backend is specified (#22217)

mgoin
authored and
jinzhen-lin
committed
[Core] Factor out common logic for MM budget calculation (#22228)

DarkLight1337
authored and
jinzhen-lin
committed
[Model] Pooling model activation supports per request control by PoolingParams (#20538)

noooop
authored and
jinzhen-lin
committed
[Docs][TPU] Highlight TPU Software version selection (#22242)

NickLucche
authored and
jinzhen-lin
committed
Migrate KimiVLImagePixelInputs to TensorSchema (#21769)

authored and
jinzhen-lin
committed
[Feature] Non-contiguous Support for FP8 Quantization (#21961)

authored and
jinzhen-lin
committed
[NVIDIA] Support Flashinfer TRT-LLM Prefill Attention Kernel (#22095)

elvischenv
authored and
jinzhen-lin
committed
[Misc] correct static type check for GroupCoordinator (#21946)

andyxning
authored and
jinzhen-lin
committed
[V0 Deprecation][TPU] Remove V1 flag check from tests (#22248)

NickLucche
authored and
jinzhen-lin
committed
Use UV_LINK_MODE=copy in Dockerfile to avoid hardlink fail (#22128)

mgoin
authored and
jinzhen-lin
committed
[CI/Build] Update flashinfer to 0.2.9 (#22233)

mgoin
authored and
jinzhen-lin
committed
[Refactor] Remove Unused Environment Variable `VLLM_NO_DEPRECATION_WARNING` (#22199)

yewentao256
authored and
jinzhen-lin
committed
[V1] port xformers backend to v1 (#21342)

TheEpicDolphin
authored and
jinzhen-lin
committed
[bugfix] fix blackwell deepep installation (#22255)

youkaichao
authored and
jinzhen-lin
committed
[CI][TPU] Fix docker clean up (#22271)

lsy323
authored and
jinzhen-lin
committed
[Bugfix] Remove faulty test for oot attention backend (#22286)

mgoin
authored and
jinzhen-lin
committed
[Bugfix] Fix 3D input passed into cutlass_scaled_mm (#22278)

mgoin
authored and
jinzhen-lin
committed
[Bugfix] Fix MoE BNB version (#22260)

jeejeelee
authored and
jinzhen-lin
committed
[Perf] Parallelize fill_bitmask to accelerate high-throughput guided decoding (#21862)

benchislett
authored and
jinzhen-lin
committed
[Bugfix] Skip dead and non-GPU nodes for Ray DP engine allocation (#22275)

ruisearch42
authored and
jinzhen-lin
committed
[Bugfix][CI/Build][ROCm] Make sure to use the headers from the build folder on ROCm (#22264)

gshtras
authored and
jinzhen-lin
committed
Upgrade FA3 for attention sink (#22313)

WoosukKwon
authored and
jinzhen-lin
committed
Increase openai-python version (#22316)

WoosukKwon
authored and
jinzhen-lin
committed
Add attention sink in attention backends (#22320)

authored and
jinzhen-lin
committed
Update transformers to `v4.55` (#21931)

authored and
jinzhen-lin
committed
Add GPT-OSS model code and config [1/N] (#22327)

WoosukKwon
authored and
jinzhen-lin
committed
[ROCm] Add attention sink to use_rocm_custom_paged_attention (#22329)

authored and
jinzhen-lin
committed
[GptOss] Add GptOss reasoning parser to support structure output (#22322)

authored and
jinzhen-lin
committed
[gpt-oss] flashinfer attention sink init (#22330)

authored and
jinzhen-lin
committed
[gpt-oss] Add openai-harmony as default dependency (#22332)

authored and
jinzhen-lin
committed
[Misc] Clean up duplicated hf overrides (#22311)

Isotr0py
authored and
jinzhen-lin
committed
[gpt-oss] Add Tool/ConversationContext classes and harmony_utils (#22340)

authored and
jinzhen-lin
committed
[gpt-oss] add model to supported models doc (#22336)

Roger Wang
authored and
jinzhen-lin
committed
[gpt-oss] Support chat completion api (#22342)

WoosukKwon
authored and
jinzhen-lin
committed
[Minor] Fix type (#22347)

WoosukKwon
authored and
jinzhen-lin
committed
[BugFix] Fix FA2 RuntimeError when sinks is provided (#22365)

LucasWilkinson
authored and
jinzhen-lin
committed
add the codes to check AMD Instinct GPU number (#22367)

zhangnju
authored and
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix fp4 layer process
jinzhen-lin
committed
[BugFix] Fix triton compile error in `kernel_unified_attention_2/3d` caused by attention sinks (#22368)

LucasWilkinson
authored and
jinzhen-lin
committed
[Bugfix] Make condition in triton kernel constexpr (#22370)

gshtras
authored and
jinzhen-lin
committed
[gpt-oss] Add loop for built-in tool call (#22374)

authored and
jinzhen-lin
committed
[gpt-oss] Enhance error msg on attention sink init (#22335)

authored and
jinzhen-lin
committed
[gpt-oss] flashinfer mxfp4 (#22339)

authored and
jinzhen-lin
committed
[v1] - Mamba1 Attention Metadata (#21249)

authored and
jinzhen-lin
committed
[Bug] Fix B200 DeepGEMM E8M0 Accuracy Issue (#22399)

yewentao256
authored and
jinzhen-lin
committed
[gpt-oss] add demo tool server (#22393)

heheda12345
authored and
jinzhen-lin
committed
[gpt-oss] fix model config with hf_config (#22401)

zyongye
authored and
jinzhen-lin
committed
Fix trtllm-gen attention env and add attention sink (#22378)

authored and
jinzhen-lin
committed
Update `flashinfer-python==0.2.10` (#22389)

mgoin
authored and
jinzhen-lin
committed
[model] Support MiniCPM-V 4.0 (#22166)

authored and
jinzhen-lin
committed
Support encoder_only attention for FlexAttention (#22273)

maxdebayser
authored and
jinzhen-lin
committed
[Attention] Support multiple attention metadata builders per kv_cache_spec + proper local attention no hybrid kv cache fix (#21588)

LucasWilkinson
authored and
jinzhen-lin
committed
[XPU]Fix `flash_attn_varlen_func` interface on xpu (#22350)

jikunshang
authored and
jinzhen-lin
committed
[Qwen3] Enable dual-chunk-attention support for Qwen3 models. (#21924)

sighingnow
authored and
jinzhen-lin
committed
[Bugfix] Fix wrong method name in Intern-S1 image processor (#22417)

DarkLight1337
authored and
jinzhen-lin
committed
Use float32 for test_completion.py (#22385)

mgoin
authored and
jinzhen-lin
committed
[Bugfix]: Fix the streaming output for function calls in the minimax (#22015)

qscqesze
authored and
jinzhen-lin
committed
[Bugfix] Add proper comparison for package versions (#22314)

syedmba
authored and
jinzhen-lin
committed
Update `hf_xet` pin to resolve hangs (#22356)

hmellor
authored and
jinzhen-lin
committed
Optimize logger init performance by using module-level constants (#22373)

skyloevil
authored and
jinzhen-lin
committed
preload heavy modules when mp method is forkserver (#22214)

lionelvillard
authored and
jinzhen-lin
committed
[gpt-oss] Convert user input to harmony format (#22402)

authored and
jinzhen-lin
committed
[Bugfix] EPLB load statistics problem (#22167)

authored and
jinzhen-lin
committed
[CI] Skip the pooling models that do not support transformers v4.55 (#22411)

noooop
authored and
jinzhen-lin
committed
[Bench] Split serve.py:main into async/async versions (#22405)

lk-chen
authored and
jinzhen-lin
committed
[Model] Switch to Fused RMS norm in Qwen2.5_VL model. (#22184)

authored and
jinzhen-lin
committed
[Frontend] Update OpenAI error response to upstream format (#22099)

msanft
authored and
jinzhen-lin
committed
[Misc] Support routing logic simulation (#21990)

authored and
jinzhen-lin
committed
feat: Add --enable-log-outputs flag for logging model generations (#20707)

mizadri
authored and
jinzhen-lin
committed
init frondend
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix scale
jinzhen-lin
committed
fix interleave
jinzhen-lin
committed
activation func test
jinzhen-lin
committed
fix activation
jinzhen-lin
committed
fix
jinzhen-lin
committed
Update csrc/moe/marlin_moe_wna16/kernel.h

jinzhen-lin
and
mgoin
committed
fix format
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix format
jinzhen-lin
committed
[Docs] Add missing dependency for docs build (#22435)

hmellor
authored and
jinzhen-lin
committed
Add H20-3e fused MoE kernel tuning configs for GLM-4.5 (#22433)

authored and
jinzhen-lin
committed
[Misc] Enhance code formatting in mxfp4.py (#22423)

WoosukKwon
authored and
jinzhen-lin
committed
[Doc] Fix link to prefix caching design (#22384)

sarckk
authored and
jinzhen-lin
committed
[Docs] Factor out troubleshooting to its own guide; add section for Ray Observability (#21578)

crypdick
authored and
jinzhen-lin
committed
[Doc] update docs for nightly benchmarks (#12022)

andrewkchan
authored and
jinzhen-lin
committed
[Docs] Update features/disagg_prefill, add v1 examples and development (#22165)

david6666666
authored and
jinzhen-lin
committed
[Core] Store only the keys for multi-modal data in P0 (#22198)

DarkLight1337
authored and
jinzhen-lin
committed
[Bugfix] Add missing `packed_modules_mapping` to `DeepseekV2ForCausalLM` (#22352)

fxmarty-amd
authored and
jinzhen-lin
committed
[Tool] Fix auto tool call (#22434)

heheda12345
authored and
jinzhen-lin
committed
[gpt-oss] Generate ResponseOutputItem from Harmony Message (#22410)

heheda12345
authored and
jinzhen-lin
committed
Fix pre-commit error in main (#22462)

WoosukKwon
authored and
jinzhen-lin
committed
[Core] Simplify mm processing cache (#22457)

DarkLight1337
authored and
jinzhen-lin
committed
[Frontend] Use engine argument to control MM cache size (#22441)

DarkLight1337
authored and
jinzhen-lin
committed
Remove `from_dict` from `SpeculativeConfig` (#22451)

hmellor
authored and
jinzhen-lin
committed
[Misc] normalize multiprocessing Queue usage (#22371)

andyxning
authored and
jinzhen-lin
committed
[ROCm] [V1] [SpecDec] Enable Speculative Decoding on ROCm V1 Engine (#21496)

tjtanaa
authored and
jinzhen-lin
committed
[PERF] Use pybase64 to more quickly decode prompt embeddings (#22469)

qthequartermasterman
authored and
jinzhen-lin
committed
Add ModelOpt Qwen3 nvfp4 support (#20101)

Edwardf0t1
authored and
jinzhen-lin
committed
Support Tensorrt-LLM MoE fp4 for low-latency (#21331)

authored and
jinzhen-lin
committed
Fix Flashinfer CUTLASS MOE Allgather (#21963)

wenscarl
authored and
jinzhen-lin
committed
[Kernel] Add support for block FP8 on SM120 (NVIDIA 5090 and RTX PRO 6000) (#22131)

0xjunhao
authored and
jinzhen-lin
committed
[Bugfix] Fix RuntimeError: Index put requires the source and destination dtypes match (#22065)

chaunceyjiang
authored and
jinzhen-lin
committed
not tie_word_embeddings for glm-4.5 and glm-4.5v (#22460)

zRzRzRzRzRzRzR
authored and
jinzhen-lin
committed
Optimize MiniCPMO mask creation with vectorized implementation (#22464)

skyloevil
authored and
jinzhen-lin
committed
Fix pre-commit (#22487)

DarkLight1337
authored and
jinzhen-lin
committed
[bugfix] Fix Llama3/4 issues caused by FlashInfer 0.2.10 (#22426)

nvpohanh
authored and
jinzhen-lin
committed
[Doc] Sleep mode documentation (#22310)

authored and
jinzhen-lin
committed
[bench] Fix benchmark/serve.py to ignore unavailable results (#22382)

lk-chen
authored and
jinzhen-lin
committed
fix topk
jinzhen-lin
committed
fix topk
jinzhen-lin
committed
[CI/Build] Fix multimodal tests (#22491)

DarkLight1337
authored and
jinzhen-lin
committed
[Misc] Begin deprecation of `get_tensor_model_*_group` (#22494)

DarkLight1337
authored and
jinzhen-lin
committed
[Misc] fix openai version (#22485)

lengrongfu
authored and
jinzhen-lin
committed
[BugFix] Don't cancel asyncio tasks directly from destructors (#22476)

njhill
authored and
jinzhen-lin
committed
[Docs] Improve API docs (+small tweaks) (#22459)

hmellor
authored and
jinzhen-lin
committed
Remove exception for Python 3.8 typing from linter (#22506)

hmellor
authored and
jinzhen-lin
committed
[gpt-oss] triton kernel mxfp4 (#22421)

zyongye
authored and
jinzhen-lin
committed
disable m_block_size_8 temporarily
jinzhen-lin
committed
fix moe_block_size_8
jinzhen-lin
committed
fix idx
jinzhen-lin
committed
fix
jinzhen-lin
committed
fix format
jinzhen-lin
committed
disable on hopper
jinzhen-lin
committed
[Benchmark] Add benchmark tool for multi turn conversations (#20267)

pliops-daniels
authored and
jinzhen-lin
committed
[gpt-oss] guard import when triton kernel is not installed (#22529)

authored and
jinzhen-lin
committed
[Docs] Rename “Distributed inference and serving” to “Parallelism & Scaling” (#22466)

crypdick
authored and
jinzhen-lin
committed
[gpt-oss] Support tool call and implement MCP tool server (#22427)

heheda12345
authored and
jinzhen-lin
committed
[BugFix] Fix IMA FlashMLA full cuda-graph and DP + Update FlashMLA (#21691)

authored and
jinzhen-lin
committed
[Misc] DeepGEMM : Avoid JIT generation in the hot-path (#22215)

authored and
jinzhen-lin
committed
[Bugfix] Update FA commit hash (#22546)

tdoublep
authored and
jinzhen-lin
committed
Skip Qwen 1 in CI because remote code is no longer compatible with Transformers (#22536)

hmellor
authored and
jinzhen-lin
committed
[Docs] fix broken links in metrics.md (#22315)

authored and
jinzhen-lin
committed
[Frontend] Add unix domain socket support (#18097)

yyweiss
authored and
jinzhen-lin
committed
Extract `CompilationConfig` from `config.py` (#22524)

hmellor
authored and
jinzhen-lin
committed
Drop flaky test_healthcheck_response_time (#22539)

russellb
authored and
jinzhen-lin
committed
[XPU] upgrade torch 2.8 on for XPU (#22300)

jikunshang
authored and
jinzhen-lin
committed
[BugFix] [P/D] Handle lookahead token count edge-case with Eagle Spec Decoding and P/D (#22317)

authored and
jinzhen-lin
committed
update use_marlin condition
jinzhen-lin
committed
Merge branch 'main' of github.com:vllm-project/vllm into marlin-mxfp4-bias
jinzhen-lin
committed
fix
jinzhen-lin
committed

Commits on Aug 12, 2025

Commits on Aug 14, 2025

Merge branch 'main' into marlin-mxfp4-bias
mgoin
committed