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
- [Performance]: Reduce load_weights synchronization overhead for DeepSeek V4 Flash weight updates on H200
- [Perf][MoE] Avoid GPU synchronization in expert ID lookups
- [Bugfix][KV Connector][ROCm] MoRIIO: accept kernel-block-split MLA/indexer KV views
- [ROCm][Kimi-K3] Mono MoE as a vLLM kernel (vllm/models/kimi_k3/amd/mono)
- [Bugfix] Gate the FA4 masked-MHA sparse prefill on prefill-query quantization, not KV cache dtype
- [Rust Frontend] Add `--always-constrain-output`
- [Performance]: Triton-based attention runs without tensor cores on Turing (SM75) since Triton 3.3, so T4 prefill is up to 15x slower than FlashInfer
- [Bug]: `top_k_per_row_decode` raises an illegal memory access for runtime topK > 8192 instead of a clean error
- [ROCm][Perf][GLM-5] Fused decode with the MonoKernel in vLLM
- [ROCm][Perf] Qwen3.8-2.4T specialized decode on MI355X (--monokernel)
- Docs
- Python not yet supported