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
- [Usage]:
- [KV Connector][Mooncake] Publish Store residency once per complete block
- [ROCm][Kernel] Add fused MoE int4_w4a16 config for E=128,N=704 on AMD Radeon AI PRO R9700
- [Bugfix][MoE] Decouple DeepEP low-latency dispatch size from max_num_batched_tokens
- [Perf][Kernel][Quantization] Triton row kernel for per-token-group FP8 quant at large token counts
- [RFC]: A standard for Specialized (MonoKernels) Decodes in vllm/models
- [Bug]: DeepSeek-V4-Pro returns wrong answers on long prompts (>~380k tokens) when --long-prefill-token-threshold=256 is set (v0.30.0)
- [Build] Reuse the CMake build directory for editable installs
- [ROCm][Bugfix] Run TheRock's ROCm SDK preload before promoting libtorch_cpu.so
- [Bug][Spec Decode]: DeepGEMM warm-up covers only the target model; speculator GEMMs JIT-load in the serving hot path
- Docs
- Python not yet supported