Repository Issues
vllm-project/vllm
A high-throughput and memory-efficient inference and serving engine for LLMs
Issues
Open
[Feature]: Optimize collectives in TP MoE case using torch.compile pass
feature requestgood first issuehelp wantedperformancetorch.compile
Has a beginner-friendly labelContributing guide available
33 comments0 reactions2 assignees
Closed
[Feature]: Fused Kernel for GPT-OSS Router
feature requestgood first issuehelp wantedstale
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
10 comments0 reactions0 assignees
Open
[Bug][RL]: Port Conflict
buggood first issuehelp wantedstaleunstale
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
17 comments2 reactions0 assignees
Open
[Feature]: Improve DCP error messages
feature requestgood first issuestale
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
5 comments2 reactions0 assignees
Open
[Bug]: AsyncTP pass has poor perf on B200
bughelp wantedkeep-opentorch.compile
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
5 comments2 reactions2 assignees
Open
[Feature]: Batch Invariant Feature and Performance Optimization
feature requestgood first issue
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
59 comments27 reactions3 assignees
Open
[Feature]: Tracking Whisper feature requests
feature requestgood first issuehelp wantedkeep-openmulti-modality
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
13 comments17 reactions0 assignees
Open
[RFC]: Limit the use of envvars in vLLM
RFCgood first issue
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
17 comments5 reactions2 assignees
Closed
[Bug]: torch.compile padding causes IMA on Hopper + DBO
bughelp wantedstaletorch.compileunstale
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
41 comments0 reactions1 assignee
Open
[Performance]: Compiled `QuantFP8.forward_native` group quantization (1, 128) slower than CUDA on H100/RTX5090
help wantedkeep-openneeds reproductionperformancetorch.compile
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
18 comments3 reactions2 assignees
Closed
[torch.compile][Performance]: Unwrap custom ops and improve fusion (Inductor and custom)
help wantedperformancetorch.compile
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
21 comments11 reactions1 assignee
Closed
[Bug]: Hangs on mulit-node setup when len(prompt) > max_model_len
buggood first issuestale
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
15 comments1 reaction1 assignee
Open
[Feature]: Support `Phi4Flash` model in V1
feature requestgood first issuehelp wantedstaleunstale
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
7 comments0 reactions0 assignees
Open
[Bug]: Add tests for all parallel sampling parameter combinations
buggood first issuekeep-openstale
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
7 comments0 reactions1 assignee
Open
[Feature]: Add Triton implementation of NVFP4 GEMM
feature requestgood first issueperformanceunstale
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
10 comments1 reaction0 assignees
Open
[Feature]: `CustomOp` cleanup
feature requestgood first issuehelp wantedkeep-openstartup-ux
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
7 comments0 reactions1 assignee
Open
[Feature]: Run performance benchmarks for multi-modal models in CI
feature requesthelp wantedkeep-openmulti-modality
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
15 comments0 reactions0 assignees
Open
[Feature]: Composite model loading using `AutoWeightsLoader` for all models
feature requestgood first issuekeep-open
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
39 comments0 reactions0 assignees
Open
[Performance]: Update Cascade Attention Heuristics for FA3
feature requestgood first issuehelp wantedstaleunstale
Why recommendedHas a beginner-friendly label · Contributing guide available
Has a beginner-friendly labelContributing guide available
20 comments0 reactions1 assignee
Open
[Feature]: Implement Concurrent Partial Prefills In V1 Engine
feature requesthelp wantedunstale
Why recommendedNo assignee yet · Has a beginner-friendly label
No assignee yetHas a beginner-friendly labelContributing guide available
17 comments0 reactions0 assignees