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
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [RFC]: Software-dequant fp8 KV cache for MLA on Ampere (sm80/sm86) — consolidate the existing pieces, a 1M-ctx field case, and a validation offer
- [Kernel] Allow steering the unquantized linear backend independently
- [ROCm][Kimi-K3] bf16 router GEMM and fused MLA output gate
- Hidden-state extraction for GLM-5.3-Flash
- [Bugfix][Benchmark] Fix bench serve result serialization
- [MM][Model] Apply HF Pixtral RoPE in attention layout
- [ROCm][Perf] Fold Q fp8 quantization into QkNormRopeKvCacheFusionPass
- [Bugfix][Spec Decode] Keep mtp.fc unquantized on quark and compressed-tensors checkpoints
- [Bug]: DeepSeek-V4.1-Flash on SM120 (RTX PRO 6000): Xid 31 illegal memory access during long prefill — deterministic with CUDA graphs enabled, flaky in eager mode (v0.30.0 + FlashInfer 0.7.0)
- [Bugfix][Metrics] Retain components for resolved quantization configs
- Docs
- Python not yet supported