Search results

[Bug]: PP=2 + GlmMoeDsa: inductor compile combined with CUDA-graph capture produces garbage output; either alone is clean (v0.24 & v0.26)

Status: Open.

#49844 In vllm-project/vllm;

· divyvasal opened on Jul 25, 2026

[Bug][KV Offload] HIT_PENDING has no deadline; a stalled write can defer requests until the client timeout

Status: Open.

#49829 In vllm-project/vllm;

· nilig opened on Jul 25, 2026

[Bug]: Cross-node Pipeline Parallelism (PP) fails with "invalid device ordinal" at PyNccl warmup on ROCm (AMD RDNA4 / R9700), reproducible on first inference even after context-corruption workaround

bugSomething isn't workingSomething isn't working

rocmRelated to AMD ROCmRelated to AMD ROCm

Status: Open.

#49826 In vllm-project/vllm;

· bertinillo opened on Jul 25, 2026

[Bug][KV Offload][P2P] Symmetric fetch can remain non-terminal after confirmed lookup hits; consumer stays deferred for 30s

Status: Open.

#49820 In vllm-project/vllm;

· nilig opened on Jul 25, 2026

[Bug]: vllm serve --help=Frontend misses docs for inherited fields from BaseFrontendArgs

bugSomething isn't workingSomething isn't working

Status: Open.

#49817 In vllm-project/vllm;

· xu1225877712 opened on Jul 25, 2026

[Feature]: per gpu gpu-memory-utilization

feature requestNew feature or requestNew feature or request

Status: Open.

#49816 In vllm-project/vllm;

· competitiveNN opened on Jul 25, 2026

[Bug]: PCP (#46570) broken for non-compress models (GLM-5.2, compress_ratio=1) — multiple crash paths

bugSomething isn't workingSomething isn't working

Status: Open.

#49810 In vllm-project/vllm;

· Leoyzen opened on Jul 25, 2026

[Bug][KV Offload][P2P] EngineCore crash reconnecting to peer: stale dead ZmqConnection remains registered

Status: Open.

#49809 In vllm-project/vllm;

· nilig opened on Jul 25, 2026

[Bug]: DeepGEMM 2.6.x UE8M0 assert - vLLM passes uninitialized FP32 scale-factor padding to the packing kernel

quantization

Status: Open.

#49783 In vllm-project/vllm;

· eous opened on Jul 25, 2026

[Bug]: gpt-oss-120b fails for ROCm on gfx950 with Triton 3.7.1

bugSomething isn't workingSomething isn't working

rocmRelated to AMD ROCmRelated to AMD ROCm

Status: Open.

#49778 In vllm-project/vllm;

· rjrock opened on Jul 25, 2026

[CI Failure]: PyTorch Compilation Unit Tests - test_dynamic_shapes_compilation

Status: Open.

#49772 In vllm-project/vllm;

· Change72 opened on Jul 25, 2026

[CI Failure]: entrypoints-integration-speech_to_text - WER tests fail through evaluate HfFolder

Status: Open.

#49771 In vllm-project/vllm;

· Change72 opened on Jul 24, 2026