Pull Requests vllm-project/vllm

[Bugfix][KV Offload] Avoid quadratic ARC batch eviction bug

#50992 opened 2026-08-04 07:47 by mindungil

[Mamba] enable prefix cache by default

#50991 opened 2026-08-04 07:40 by ZJY0516

[Bugfix][Multimodal] Validate dynamic video sampling metadata bug multi-modality

#50990 opened 2026-08-04 07:32 by zcxGGmu

[Quantization][AutoRound] Add NVFP4 linear support quantization

#50988 opened 2026-08-04 07:21 by wimCCC

[Bugfix][DeepSeek V4] Fix fused MTP RMSNorm dtype mismatch bug deepseek

#50987 opened 2026-08-04 07:06 by dongjibin1996

[Bugfix][Tool Calling] Validate Qwen3 tool names in Parser Engine bug tool-calling qwen

#50986 opened 2026-08-04 07:03 by c5566b

improve OOM error messages for low-VRAM scenarios

#50985 opened 2026-08-04 06:49 by YifanLi3

[Bugfix][Mooncake] Report failed remote KV loads to the scheduler bug kv-connector

#50984 opened 2026-08-04 06:40 by BruceLoveDecimal

[Bugfix] Preserve the audio duration-limit error through the PyAV fallback bug multi-modality

#50983 opened 2026-08-04 06:33 by ieaves

[Bugfix][MoE] Fall back to torch for unsupported expert counts in topk_softplus_sqrt bug

#50982 opened 2026-08-04 06:32 by AiKiAi-stack

[MISC][Bench] refactor throughput and reuse serve's get samples performance

#50981 opened 2026-08-04 06:24 by JaredforReal

[HPC Attention Backend] hpc attention backend support bf16 kv cache with fp8 weight

#50980 opened 2026-08-04 06:15 by thisjiang

[Bugfix][Sampler] Keep exactly k tokens in top-k with tied logits bug

#50979 opened 2026-08-04 06:11 by liuxiang-sudo

[Bugfix][Reasoning] Prevent Gemma4 pre-reasoning text from leaking into content bug tool-calling

#50978 opened 2026-08-04 06:07 by jungjiyu

profiler: add PrivateUse1 activity support for custom backends

#50977 opened 2026-08-04 06:07 by dmholtz

Handle unavailable ROCm GCN architecture fallback rocm intel-gpu quantization

#50976 opened 2026-08-04 06:01 by xwu-intel

[Frontend][Responses API] Support assistant message `phase` frontend ci/build gpt-oss

#50975 opened 2026-08-04 05:58 by duz52

[compile] Remove layer_name from unified_kv_cache_update

#50973 opened 2026-08-04 05:42 by bandham-manikanta

[Bugfix][Core] Add VLLM_FORCE_FP32_ALL_REDUCE env var to fix PCIe non-english logit drift (#40725) bug

#50972 opened 2026-08-04 05:27 by Vinay12345-neutron

[Dockerfile][KVConnector] Add LMCache support with native C++ compilation to Dockerfile documentation performance frontend speculative-decoding ci/build cpu kv-connector nvidia rust kimi k3

#50971 opened 2026-08-04 05:16 by Snagwick

Bump the minor-update group across 1 directory with 171 updates rocm ci/build cpu nvidia dependencies

#50970 opened 2026-08-04 05:09 by dependabot[bot]

[BugFix] RayExecutorV2: choose the TCPStore port on the node that binds it bug

#50969 opened 2026-08-04 05:03 by aoshen02

[EPLB]supports kimi k3 eplb kimi k3

#50967 opened 2026-08-04 04:52 by chaunceyjiang

[Bugfix] Fix get_open_port() livelock on DP-reserved ports and cover get_open_ports_list bug

#50965 opened 2026-08-04 04:34 by aoshen02

[Bugfix][Gemma4] Align enable_thinking default with chat template bug tool-calling

#50964 opened 2026-08-04 04:34 by DavidDiao

[Bugfix][Model] GraniteMoeHybrid: skip other ranks' weights under pipeline parallelism bug

#50962 opened 2026-08-04 04:15 by nikJ13

[Bugfix] Fix ZMQ port TOCTOU race in shm_broadcast MessageQueue bug

#50960 opened 2026-08-04 04:04 by aoshen02

[Bugfix][Model] Gemma3n/Gemma4: pad variable-length audio batches bug multi-modality

#50958 opened 2026-08-04 03:53 by TrainToGPB

[Bugfix] Preserve Qwen3 structured output content bug frontend tool-calling qwen

#50955 opened 2026-08-04 03:33 by jeffdecoste

[Bugfix] Pad missing draft rows in the grammar bitmask bug structured-output

#50952 opened 2026-08-04 03:14 by he-yufeng