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
Add a CodeTriage badge to vllm
Help out
- Issues
- [ROCm][Perf] Use AITER single-launch KDA decode in GLM-5.3-Flash
- allocate_kv_cache raises StopIteration when a PP rank owns none of a KV group's layers
- [Perf][MLA] Cap TRITON_MLA decode num_kv_splits for single-request batches on ~100 KiB-smem GPUs
- [RFC]: Lossless 12-bit storage for BF16 dense weights to speed up memory-bound decode
- [Perf] Tuned batch-invariant fused MoE configs on Blackwell
- [Bugfix][Reasoning] Hunyuan A13B: detect reasoning end from token ids so structured output applies
- [Perf][Kernel] Reduce shared-memory staging for per-tensor FP8 KV dequantization
- [Docs] Add GLM-5.3-Flash to encoder CUDA graph supported models
- [Perf][Beam] Fuse FP8 KV-cache insertion and query quantization into QK-norm/RoPE
- [Bug]: MoE models crash at startup on torch 2.11.0 with default VLLM_USE_LAYERNAME=1 (inductor assert __obj_flatten__ on hoisted LayerName)
- Docs
- Python not yet supported