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
- [Bugfix][Kernel] Make Marlin MoE route alignment deterministic
- [Performance]: Official FP8 checkpoint ~40% slower at bs=1 than runtime fp8 quant of same model (Qwen3-4B, RTX 4090, v0.24.0)
- [Bugfix][Scheduler] Let pooling chunked prefill use full context
- [Bug][ROCm] Triton kernel_paged_attention_2d illegal memory access (hipErrorLaunchFailure) on gfx950 (MI350X) under long-context decode
- [RFC][Feature]: First-class session id for conversation aware routing
- [ROCm] Don't re-quantize bf16 MLA kv_b_proj to fp4/fp8 for GLM MoE DSA
- [Bugfix] Fix int32 offset overflow in rejection sampler kernels
- [Kernel] Marlin MoE: allow overriding block_size_m via VLLM_MARLIN_MOE_BLOCK_SIZE_M
- Optimize ppc64le vLLM runtime image size
- [Performance]: Triton attention drops long-context decode from 3D split-KV to 2D at batch ≥ ~12, ~doubling ITL
- Docs
- Python not yet supported