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 supported51 Subscribers
Add a CodeTriage badge to vllm
Help out
- Issues
- Add FlashInfer fused decode for Qwen GDN on sm120
- [Bugfix] Fix DFlash vocabulary module initialization
- EAGLE/MTP prefix-cache last-block drop causes a 1,648-token recompute per hit on one hybrid Qwen3.8 GDN layout — ~30-40% batch throughput loss on prefix-reusing workloads with speculative decoding
- [Bugfix][Kernel] NVFP4 KV cache: fix block_size/layout detection for physically shaped HND caches
- [Core] Skip common-prefix block scan for single requests
- [Model][Quant] DeepSeek-V4: opt-in lossless MXFP4->block-FP8 expert dequant (Hopper prefill speedup)
- [Kernel][Perf] Tune fused_moe FP8 config for Qwen3-VL on L40S (+8% at batch 4-512)
- [Bug]: Named shared memory may be closed before peer ranks open it
- [Core] Add default-off preemption_victim="lcf" (least-computed-first) victim policy
- [DCP][Mooncake] Harden hybrid prefix-cache external hits
- Docs
- Python not yet supported