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
- [Bugfix][KV Connector][Mooncake] Report Mooncake Store reset failures
- [Bugfix][LoRA] Respect moe_backend and guard fully-sharded LoRAs in MoE selection
- [XPU][LoRA] Support FP8 and W4A16 modular MoE experts
- [Bug]: disable_hybrid_kv_cache_manager=True crashes engine startup on hybrid mamba models — ValueError: Failed to promote local KV cache specs to one unified type
- [Bug]: fp8 kernel selection discards per-kernel rejection reasons — quantization="fp8_per_channel" surfaces as a misleading "Requires SM89+ for Cutlass FP8" with no indication of what actually failed
- [Bugfix][Spec Decode] Serialize batch queue for spec decode + PP
- [Bug]: DFlash speculative decoding with gpt-oss target crashes with FA3 illegal memory access
- [ROCm][Attention] Honor VLLM_ROCM_USE_AITER=1 by ranking ROCM_AITER_FA above ROCM_ATTN
- [Bugfix][Kernel] MXFP4 cutlass MoE: fix experts quant out-of-bounds read for expert counts not divisible by 4 or 16
- [Bug]: DFlash drafter CUDA graph capture fails with FLEX_ATTENTION backend (seq_lens_cpu_upper_bound is None)
- Docs
- Python not yet supported