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][Spec Decode][Quantization] Handle quantized qkv_proj in DFlash fused-KV buffers
- [Bug]: DeepSeek-V4-Flash TP on H100: --max-num-batched-tokens >= 24576 crashes EngineCore in fused qnorm/rope/kv-insert op; allocation invisible to memory profiler
- [Bug]: Hybrid block-size alignment is skipped on pipeline-parallel ranks that own no attention layer
- [Bugfix] Fix NVIDIA DeepSeek V4 mHC warmup
- [Model] Add DeepGrove Maple (MapleForCausalLM)
- [Security] Include cache_salt in HF3FS external cache keys
- [Bugfix][Core] Align hybrid block sizes across PP stages
- EAGLE/MTP block drop + prefix caching is untested for hybrid models with ≥3 attention groups (DeepSeek-V4-Flash + DSpark lands there)
- [Bugfix][MRv1] Decouple async Mamba align D2H counts from InputBatch row shifts (#51571)
- [Bug]: qwen3_xml tool parser consumes `</think>`, merging reasoning into `content` with no way to split it
- Docs
- Python not yet supported