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] Fix per-group prefix-hit divergence for hybrid Mamba + KV connector
- [Bugfix][CPU][RISC-V] Use explicit rounding mode for vfcvt float-to-int conversions
- [Bug]: DeepSeek V4/V32 parser can unwrap arguments using the wrong tool's schema when parallel tool calls share identical raw argument text
- [Bugfix] Handle E8M0 block scales in CUTLASS and Triton FP8 linear kernels
- [Bug]: SM120 CUTLASS blockwise FP8 GEMM rejects N % 128 != 0 (Invalid status; kv_a_proj N=576 in DeepSeek shapes)
- [RFC][Frontend/CLI]: Offline prefix-cache workload analyzer
- VLLM_ENGINE_READY_TIMEOUT_S default (600s) insufficient for large MoE + cold FlashInfer JIT cache on new GPU archs (GB10 / sm_121a)
- [RFC]: PyTorch vLLM correctness audit
- [Bug]: DeepSeek-V4 sparse-SWA decode passes flashinfer 0.6.14-only kwargs, but the pin is 0.6.13
- [Performance]: Marlin MoE block_size_m heuristic over-selects a large M-tile for expert-parallel decode (up to ~30% TPOT on gpt-oss)
- Docs
- Python not yet supported