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
- [Performance]: MoE-LoRA one-shot kernel: rank offsets defeat load vectorization (w2 at 14% DRAM, 92% excess sectors); ~2x with bit-identical output
- [Feature] Add optional GPU timing to iteration metrics
- [Feature] Add workload snapshots to iteration metrics
- [Perf][Model Loader] DeepSeek V4.1 Flash mixed RunAI loading
- [Perf][Model] Qwen4Exp QSA DCP: follow-ups to #59279 (combine, workspace buffers, MTP decode path)
- [RFC]: DFlash2 speculative decoding performance on ROCm (MI355X)
- [Bugfix][LoRA] Account for CLIP CLS rows in LLaVA tower mappings
- [Bug]: fused_recurrent_gated_delta_rule allocates final_state for one sequence but writes it for the whole batch, resulting in out-of-bounds writes
- [Bug]: Scheduler livelock with multi-module MTP + Mamba align prefix caching when prompt_len % block_size < num_speculative_tokens
- [Observability] Add tool-calling telemetry to HTTP render endpoints
- Docs
- Python not yet supported