Llama2 API

llama-cpp-python FastAPI Server Template

This template installs the Python wrapper of llama.cpp from llama-cpp-python and runs a FastAPI-based Python server. The server can be utilized as a drop-in replacement for the OpenAI API. This means that code compatible with the ChatGPT API will also be compatible with this self-hosted version.

Features

• Compatibility: Compatible with code designed for the OpenAI API, making it easy to transition to self-hosting.

• Pre-Converted and Quantized Model: Configured to fetch a pre-converted and quantized llama code instruct model from TheBloke on Hugging Face. However, you have the flexibility to use other compatible models by editing the configuration.

• Interactive Documentation: Upon starting the server, open the deployment URL and append /docs to access a Swagger-like API documentation. Here, you can view and test all available endpoints easily.

• Customizable Token Settings: By default, the max_token setting of this model is restrictive to support less capable hardware. You can adjust this setting by navigating into the installed folder /venv/lib/python3.8/site-packages/llama_cpp/server/app.py and editing line 412 in the following manner:

max_tokens_field = Field(
    default=400, ge=1, description="The maximum number of tokens to generate."
)

Usage

• Installation: Clone this repository and install the dependencies.

• Configuration: Edit the configuration to specify your desired model or settings.

• Start the Server: Execute the command to start the server.

• Access Documentation: Open the deployment URL in your browser and append /docs to access the interactive API documentation.

Note

Please be aware that the default configuration might not suit your specific requirements. Feel free to customize the settings according to your needs.