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
Add a CodeTriage badge to vllm
Help out
- Issues
- DFlash speculator `capture()` logs "Capturing model ..." even when nothing is captured — makes drafter-capture state unobservable
- [Bug]: In a Python environment, when requesting a model deployed with vLLM and the output mode is streaming, there is a low probability that the connection is established but the model does not receive the request.
- [Attention][Spec Decode] Enable varlen Mamba2 decode for adaptive verification
- [Discussion] Tiered SWA/Mamba checkpointing (HBM tail + store periodic) + recompute backfill for divergent hybrid prefix hits
- [Bugfix] Reject shape-aliased prefills in uniform-decode classification
- [Bugfix][Compilation] Guard empty splitting ops before early return
- DFlash draft model does not inherit the target's RoPE layout (is_neox_style)
- [XPU][MoE] Tune Triton fused MoE for Intel XPU
- [KV Connector] Add scheduler context for lazy block access
- [Bugfix][CPU][RISC-V] Return zero below exp clamp bound
- Docs
- Python not yet supported