vllm
https://github.com/vllm-project/vllm
Python
A high-throughput and memory-efficient inference and serving engine for LLMs
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported62 Subscribers
Add a CodeTriage badge to vllm
Help out
- Issues
- [Bug][ROCm][MoRIIO] In READ mode, a stopped prefill's KV cache stays allocated until the decode that read it exits, so the prefill cannot restart on the same GPU (in WRITE mode, the same happens to a stopped decode)
- [Bug]: --convert classify with --runner generate crashes engine startup with a bare AssertionError (pooler_config is not None)
- [DeepSeek-V4-P2-01] Make sparse prefill request-local in BI mode
- [DeepSeek-V4-P2-02] Stabilize C128 sparse metadata for graph replay
- [Deepseek-V4-P2-03] Decode index kernels
- [DeepSeek-V4-P2-04] Reuse DeepSelect for deterministic sparse top-k
- [DeepSeek-V4-P2-05] Opt into sparse FlashMLA decode
- [Bug]: EPLB async expert rearrangement crashes engine with AssertionError in fused MoE shared_experts (GLM-5.3-Flash, v0.31.0, TP8/EP8 + MTP)
- [Bugfix][Benchmark] Fix fused_topk import in CUTLASS MoE benchmarks
- [Tracking Feature]: Support vLLM with Python 3.15
- Docs
- Python not yet supported