Files
sglang/docs/flashinfer.md
2024-01-15 16:12:57 -08:00

24 lines
610 B
Markdown

## Flashinfer Mode
[flashinfer](https://github.com/flashinfer-ai/flashinfer) is a kernel library for LLM serving.
It can be used in SGLang runtime to accelerate attention computation.
### Install flashinfer
Note: The compilation can take a very long time.
```bash
git submodule update --init --recursive
pip install 3rdparty/flashinfer/python
```
### Run a Server With Flashinfer Mode
Add `--model-mode flashinfer` argument to enable flashinfer when launching a server.
Example:
```bash
python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000 --model-mode flashinfer
```