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
- [Bug]: Qwen4Exp is gated off on XPU, but most of what the gate hides is not XPU-specific
- [Bugfix] Reap orphaned /dev/shm region files via flock-based liveness
- [Bug]: Identity entry in packed_modules_mapping silently disables layer_quant_config (Quark)
- Fix DSML leaking for DeepSeek-v4 models
- [Bug]: Speculative kv_cache_dtype=fp8 crashes startup on hybrid GDN models - FlashInfer "Unrecognized dtype: auto", then native crash on SM89 with explicit target dtype
- [RFC]: Custom all-reduce for XPU
- [Misc] Add trace context to JSON logs
- [Feature] Fix KV cache group sizing for arbitrary hybrid models
- [Bugfix] Include multimodal limits in the model compile-cache hash
- [Perf]: DeepSeek-V4-Flash TP4 on 4xH800: v0.28.0 needs ~4 GiB/GPU more non-KV memory than v0.26.0; my earlier TTFT-regression claim was caused by VLLM_USE_BREAKABLE_CUDAGRAPH=0 on my side (retracted)
- Docs
- Python not yet supported