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
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Feature] Per-row candidate IDs for prefill token scoring (M2 of #56860)
- [Performance][ROCm][gfx1100] compressed-tensors silently enables fp8 KV cache, which is far slower than bf16 for paged decode attention on RDNA3
- [Perf][Attention] Source Triton metadata builder geometry from kv_cache_spec
- [MoE] Allow a fused-MoE kernel to be rebuilt around resident weights
- [Core][Model Loader] Add sleep/wake/reload lifecycle hooks to model loader
- [ROCm][Perf] W4A16: widen skinny GEMM LDS activation loads to b128
- [Bug]: `VLLM_BATCH_INVARIANT=1` is not batch-invariant for fused-MoE experts with no warning
- [Bugfix] Bound request seeds to the int64 range
- [Bug][ROCm] AiterExperts + DeepEPHT/Mori PrepareAndFinalize: illegal memory access at world_size=2 (gfx942)
- 1M-context models on MRV2: `warmup_kernels()` dummy-execute leaves the engine dead at startup/capture on ROCm — skipping that call is the only thing that fixed it for us
- Docs
- Python not yet supported