Massaki Archambault 3b41d81357 | ||
---|---|---|
librechat | ||
litellm | ||
ollama | ||
.env | ||
.gitignore | ||
README.md | ||
docker-compose.yml |
README.md
librechat-mistral
A quick prototype to self-host LibreChat with Mistral, and a OpenAI-like api provided by LiteLLM on the side.
Currently setup to run on an AMD GPU (RX 7xxx series), although the deployment could be adapted for Nvidia or other AMD GPUS
Goals
- Deploy a ChatGPT Clone for daily use.
- Deploy an OpenAI-like API for hacking on Generative AI using well-supported libraries.
- Use docker to prepare for an eventual deployment on a container orchestration platform like Kubernetes.
Getting started
Prerequisites
- Linux (WSL2 is untested)
- An AMD 7xxx series GPU (technically optional, Ollama will fallback to using the CPU but it will be very slow. Other GPUS are supported but the deployment must be modified to use them)
- docker
- docker-compose
Steps
- Clone the repo
- Run
docker-compose up
. Wait for a few minutes for the model to be downloaded and served. - Browse http://localhost:3080/
- Create an admin account and start chatting!
The API along with the APIDoc will be available at http://localhost:8000/
Configuring additional models
SASS services
Read: https://docs.librechat.ai/install/configuration/dotenv.html#endpoints
TL:DR
Let say we want to configure an OpenAI API key.
- Open the .env file.
- Uncomment the line
# OPENAI_API_KEY=user_provided
. - Replace
user_provided
with your API key. - Restart LibreChat
docker-compose restart librechat
.
Refer to the LibreChat documentation for the full list of configuration options.
Ollama (self-hosted)
Browse the Ollama models library to find a model you wish to add. For this example we will add mistral-openorca
- Open the docker-compose.yml file.
- Find the
ollama
service. Find thecommand:
option under the ollama sevice. Append the name of the model you wish to add at the end of the list (eg:command: mistral mistral-openorca
). - Open the litellm/config.yaml file.
- Add the following a the end of the file, replace {model_name} placeholders with the name of your model
- model_name: {model_name}
litellm_params:
model: ollama/{model_name}
api_base: http://ollama:11434
eg:
- model_name: mistral-openorca
litellm_params:
model: ollama/mistral-openorca
api_base: http://ollama:11434
- Open the librechat/librechat.yaml file.
- In our case, mistral-openorca is a variation of mistral-7b so we will group it with the existing Mistral endpoint. Refer to the LibreChat documentation if you wish to organize your new model as a new Endpoint.
models:
default: ["mistral-7b"]
becomes:
models:
default: ["mistral-7b", "mistral-openorca"]
- Restart the stack
docker-compose restart
. Wait for a few minutes for the model to be downloaded and served.
Architecture components
- LibreChat is a ChatGPT clone with support for multiple AI endpoints. It's deployed alongside a MongoDB database and Meillisearch for search. It's exposed on http://localhost:3080/.
- LiteLLM is an OpenAI-like API. It is exposed on http://localhost:8000/ without any authentication by default.
- Ollama manages and serve the local models.
TODO
At the time of this project, I only had access to a Linux machine with an AMD RX 7800XT GPU. I would like to include support for Windows and/or Nvidia GPUs when I get the chance.