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][Ray] Strip brackets from IPv6 node IP for MessageQueue connect_ip
- [Bug]: NVFP4 MoE monolithic backend (trtllm_fp4_block_scale_moe) produces semantically corrupt output for non-gated ReLU² MoE (Nemotron-3-Nano) on SM100; flashinfer_cutlass is correct on the same GPU
- [Bug]: MRv2: --enable-prefix-caching (mamba_cache_mode 'all') + dspark spec decode crashes at startup — prev_last_scheduled_idx never passed by MambaHybridAttnMetadata
- [Perf][DSV4] Prefer Humming over Marlin for MXFP4 MoE on SM90 (+19.9% throughput)
- [Bug]: Silent generation stall (Avg generation throughput drops to 0.0, no errors, /health and /v1/chat/completions still return 200) on Qwen3.5-397B-A17B / Qwen3.5-397B-A17B-FP8 / Qwen3.5-122B-A10B across v0.18.0, v0.19.0, v0.25.1
- [CI Perf] Trimming the tests for the hybrid models
- [bot] Enable google/gemma-4-E2B-it-qat-mobile-ct inference on Intel XPU
- [Feature]: Native SM120 backend for Quark/OCP MXFP6 (Dense and Qwen3.5 MoE)
- [hw-agnostic] Adding embedding layer and lm_head
- [Refactor] Refactor EPD
- Docs
- Python not yet supported