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 supported3 Subscribers
Add a CodeTriage badge to vllm
Help out
- Issues
- [Bug]: Gemma2 supports 8192 context with sliding window, but vllm only does 4196 or fails if try 8192
- [Usage]: vllm openai api server never ends in most cases
- [Bug]: AsyncEngineDeadError: Task finished unexpectedly with Gemma2 9B
- [Bug]: Guided decoding with Phi-3-small crashes
- [Usage]: Is there a way to make the results of two different calls to VLLM with temperature > 0 consistent?
- [Bug]: safetensor format support for Mistral-7B-v0.3
- [Installation]: pip install -e .
- [Bug]: got RuntimeError: Triton Error [CUDA]: device kernel image is invalid when running deepseek-v2
- [Usage]: The GPU KV cache usage is still small (below 20% most of time) though I turned the enable_prefix_caching on as well as modified the block_size from 16 to 32. How could I numerally increase the KV cache for the Radix prefix caching?
- c4ai-command-r-plus on 16GPUs
- Docs
- Python not yet supported