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
- [Bug]: DeepSeek-V4.1-Flash teacher-forced logprob shift on tool/structured text between nightlies dc36fcce and cd10ed6f (SM103, fp8_ds_mla)
- [Bugfix] Stop re-forcing thinking-budget end tokens once the forced end sequence is complete
- [Bugfix][KV Offload] Extend partial-tail offload to blocks_per_chunk > 1 (align-mode Mamba under MTP/EAGLE)
- [Bug]: TP=1 silent engine wedge on Qwen3.8 GDN-hybrid MoE (MTP4, NVFP4-KV, FlashInfer, SM120) — frozen generation counter, running>0 waiting=0, health 200, no self-recovery on v0.27.2rc1.dev77
- [Feature]: instruction-preserving truncation for `/v1/chat/completions`, and document `truncate_prompt_tokens` semantics
- [Bugfix][Frontend] Preserve DeepSeek's unclosed final tool parameter
- [Bugfix] DiffusionGemma: `diffusion_samples` to fan one seeded canvas out into n noise draws
- [Spec Decode][Qwen3.8] Enable fused draft metadata updates for the QSA side cache
- [Bugfix] Ignore DeepSeek V4 thinking sampling overrides
- [CI] Retry transient tokenizer loads in detokenizer tests
- Docs
- Python not yet supported