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]: Engine startup on DGX Spark (GB10, unified memory) collapses host memory — NV_ERR_NO_MEMORY while MemAvailable reports ~22 GiB
- [Bug]: GLM-5.3-Flash long-decode degeneration after accumulated reasoning decode
- [Bug]: Qwen/Qwen1.5-MoE-A2.7B-Chat produces degenerate output with torch.compile (VLLM_COMPILE) but correct output without it on vLLM 0.28.0 (H100, torch 2.13.0+cu130); CUDA graphs, compile cache and custom_ops make no difference
- [XPU] Dispatch nn.LayerNorm to fused SYCL kernel via CustomOp
- hybrid Mamba/GDN + speculative decoding: Xid 31 in the align mamba path (fault address below the CUDA segment)
- [Bug]: DeepSeek-V4.1-Flash + DSpark speculative decoding — illegal memory access in SM120 sparse-MLA prefill on long prompts
- fix: use dtype instead of deprecated torch_dtype
- [Metrics] Say what vllm:prompt_tokens and vllm:request_prompt_tokens count
- [KV Connector][Mooncake] Expose transfer stats via Prometheus
- [Security] Size Whisper audio dummies from processor chunk_length
- Docs
- Python not yet supported