61 lines
2.1 KiB
Markdown
61 lines
2.1 KiB
Markdown
# Config LLamaSharp
|
|
|
|
BotSharp contains LLamaSharp plugin that allows you to run local llm models. To use the LLamaSharp, you need to config the BotSharp project with few steps.
|
|
|
|
## Install LLamaSharp Backend
|
|
|
|
Before use LLamaSharp plugin, you need to install one of the LLamaSharp backend services that suits your environment.
|
|
|
|
- [`LLamaSharp.Backend.Cpu`](https://www.nuget.org/packages/LLamaSharp.Backend.Cpu): Pure CPU for Windows & Linux. Metal for Mac.
|
|
- [`LLamaSharp.Backend.Cuda11`](https://www.nuget.org/packages/LLamaSharp.Backend.Cuda11): CUDA 11 for Windows and Linux
|
|
- [`LLamaSharp.Backend.Cuda12`](https://www.nuget.org/packages/LLamaSharp.Backend.Cuda12): CUDA 12 for Windows and Linux
|
|
|
|
**Please install the same version of LLamaSharp Backend with the LLamaSharp in BotSharp.Plugin.LLamaSharp.csproj.**
|
|
|
|
|
|
|
|
```shell
|
|
# move to the LLamaSharp Plugin Project
|
|
$ cd src/Plugins/BotSharp.Plugin.LLamaSharp
|
|
# Install the LLamaSharp Backend
|
|
$ dotnet add package LLamaSharp.Backend.Cpu --version 0.9.1
|
|
```
|
|
|
|
## Download and Config Local LLM Models
|
|
|
|
LLamaSharp supports many LLM Models like LLaMA and Alpaca. Download the `gguf` format models and save them in your machine.
|
|
|
|
We will use a [Llama 2](https://huggingface.co/TheBloke/llama-2-7B-Guanaco-QLoRA-GGUF) model in this tutorial.
|
|
|
|
After downloading the model, open the `src/WebStarter/appsettings.json` file to config the LLamaSharp models. Set the `LlmProviders` and `LlamaSharp` fields to correct settings as your computer. For example:
|
|
|
|
```json
|
|
{
|
|
...,
|
|
"LlmProviders": [
|
|
...,
|
|
{
|
|
"Provider": "llama-sharp",
|
|
"Models": [
|
|
{
|
|
"Name": "llama-2-7b.Q2_K.gguf",
|
|
"Type": "chat"
|
|
}
|
|
]
|
|
},
|
|
...
|
|
],
|
|
...,
|
|
"LlamaSharp": {
|
|
"Interactive": true,
|
|
"ModelDir": "/Users/wenwei/Desktop/LLM",
|
|
"DefaultModel": "llama-2-7b.Q2_K.gguf",
|
|
"MaxContextLength": 1024,
|
|
"NumberOfGpuLayer": 20
|
|
},
|
|
...
|
|
}
|
|
```
|
|
|
|
For more details about LLamaSharp, visit [LLamaSharp - GitHub](https://github.com/SciSharp/LLamaSharp).
|