Serge - LLaMA made easy 🦙

Serge is a chat interface crafted with llama.cpp for running GGUF models. No API keys, entirely self-hosted!

• 🌐 SvelteKit frontend
• 💾 Redis for storing chat history & parameters
• ⚙️ FastAPI + LangChain for the API, wrapping calls to llama.cpp using the python bindings

🎥 Demo:

demo.webm

⚡️ Quick start

🐳 Docker:

docker run -d \
    --name serge \
    -v weights:/usr/src/app/weights \
    -v datadb:/data/db/ \
    -p 8008:8008 \
    ghcr.io/serge-chat/serge:latest

🐙 Docker Compose:

services:
  serge:
    image: ghcr.io/serge-chat/serge:latest
    container_name: serge
    restart: unless-stopped
    ports:
      - 8008:8008
    volumes:
      - weights:/usr/src/app/weights
      - datadb:/data/db/

volumes:
  weights:
  datadb:

Then, just visit http://localhost:8008, You can find the API documentation at http://localhost:8008/api/docs

🌍 Environment Variables

The following Environment Variables are available:

Variable Name	Description	Default Value
SERGE_DATABASE_URL	Database connection string	sqlite:////data/db/sql_app.db
SERGE_JWT_SECRET	Key for auth token encryption. Use a random string	uF7FGN5uzfGdFiPzR
SERGE_SESSION_EXPIRY	Duration in minutes before a user must reauthenticate	60
NODE_ENV	Node.js running environment	production

🖥️ Windows

Ensure you have Docker Desktop installed, WSL2 configured, and enough free RAM to run models.

☁️ Kubernetes

Instructions for setting up Serge on Kubernetes can be found in the wiki.

🧠 Supported Models

Category	Models
Alfred	40B-1023
BioMistral	7B
Code	13B, 33B
CodeLLaMA	7B, 7B-Instruct, 7B-Python, 13B, 13B-Instruct, 13B-Python, 34B, 34B-Instruct, 34B-Python
Codestral	22B v0.1
Gemma	2B, 1.1-2B-Instruct, 7B, 1.1-7B-Instruct
Gorilla	Falcon-7B-HF-v0, 7B-HF-v1, Openfunctions-v1, Openfunctions-v2
Falcon	7B, 7B-Instruct, 40B, 40B-Instruct
LLaMA 2	7B, 7B-Chat, 7B-Coder, 13B, 13B-Chat, 70B, 70B-Chat, 70B-OASST
LLaMA 3	11B-Instruct, 13B-Instruct, 16B-Instruct
LLaMA Pro	8B, 8B-Instruct
Med42	70B
Medalpaca	13B
Medicine	Chat, LLM
Meditron	7B, 7B-Chat, 70B
Meta-LlaMA-3	8B, 8B-Instruct, 70B, 70B-Instruct
Mistral	7B-V0.1, 7B-Instruct-v0.2, 7B-OpenOrca
MistralLite	7B
Mixtral	8x7B-v0.1, 8x7B-Dolphin-2.7, 8x7B-Instruct-v0.1
Neural-Chat	7B-v3.3
Notus	7B-v1
Notux	8x7b-v1
Nous-Hermes 2	Mistral-7B-DPO, Mixtral-8x7B-DPO, Mistral-8x7B-SFT
OpenChat	7B-v3.5-1210
OpenCodeInterpreter	DS-6.7B, DS-33B, CL-7B, CL-13B, CL-70B
OpenLLaMA	3B-v2, 7B-v2, 13B-v2
Orca 2	7B, 13B
Phi 2	2.7B
Phi 3	mini-4k-instruct, medium-4k-instruct, medium-128k-instruct
Python Code	13B, 33B
PsyMedRP	13B-v1, 20B-v1
Starling LM	7B-Alpha
SOLAR	10.7B-v1.0, 10.7B-instruct-v1.0
TinyLlama	1.1B
Vicuna	7B-v1.5, 13B-v1.5, 33B-v1.3, 33B-Coder
WizardLM	2-7B, 13B-v1.2, 70B-v1.0
Zephyr	3B, 7B-Alpha, 7B-Beta

Additional models can be requested by opening a GitHub issue. Other models are also available at Serge Models.

⚠️ Memory Usage

LLaMA will crash if you don't have enough available memory for the model

💬 Support

Need help? Join our Discord

🧾 License

Nathan Sarrazin and Contributors. Serge is free and open-source software licensed under the MIT License and Apache-2.0.

🤝 Contributing

If you discover a bug or have a feature idea, feel free to open an issue or PR.

To run Serge in development mode:

git clone https://github.com/serge-chat/serge.git
cd serge/
docker compose -f docker-compose.dev.yml up --build

The solution will accept a python debugger session on port 5678. Example launch.json for VSCode:

{
    "version": "0.2.0",
    "configurations": [
        {
            "name": "Remote Debug",
            "type": "python",
            "request": "attach",
            "connect": {
                "host": "localhost",
                "port": 5678
            },
            "pathMappings": [
                {
                    "localRoot": "${workspaceFolder}/api",
                    "remoteRoot": "/usr/src/app/api/"
                }
            ],
            "justMyCode": false
        }
    ]
}