30 lines
1.1 KiB
Markdown
30 lines
1.1 KiB
Markdown
# vLLM
|
|
|
|
vLLM is a fast and easy-to-use library for LLM inference and serving with an OpenAI-compatible API. Use it to run large language models on your own hardware.
|
|
|
|
## Dependencies
|
|
|
|
None, but requires a GPU node in your cluster.
|
|
|
|
## Configuration
|
|
|
|
Key settings configured through your instance's `config.yaml`:
|
|
|
|
- **model** - Hugging Face model to serve (default: `Qwen/Qwen2.5-7B-Instruct`)
|
|
- **maxModelLen** - Maximum sequence length (default: `8192`)
|
|
- **gpuProduct** - Required GPU type (default: `RTX 4090`)
|
|
- **gpuCount** - Number of GPUs to use (default: `1`)
|
|
- **gpuMemoryUtilization** - Fraction of GPU memory to use (default: `0.9`)
|
|
- **domain** - Where the API will be accessible (default: `vllm.{your-cloud-domain}`)
|
|
|
|
## Access
|
|
|
|
After deployment, the OpenAI-compatible API will be available at:
|
|
- `https://vllm.{your-cloud-domain}/v1`
|
|
|
|
Other apps on the cluster (such as Open WebUI) can connect internally at `http://vllm-service.llm.svc.cluster.local:8000/v1`.
|
|
|
|
## Hardware Requirements
|
|
|
|
This app requires a GPU node in your cluster. Adjust the `gpuProduct`, `gpuCount`, and memory settings to match your available hardware.
|