LLM Playground is a versatile environment for experimenting with different large language models (LLMs). It facilitates basic evaluation and comparisons directly in your browser, without the need to set up projects or write code in Jupyter notebooks. This tool supports a variety of LLMs, including OpenAI models, through configurable endpoints.

• Flexible Configuration: Use environmental variables, a settings YAML file, or a .env file.
• Support for Multiple Vendors: Compatible with OpenAI protocol (you can configure other providers by setting up an LiteLLM Proxy).
• Easy to Use: Designed for straightforward setup and minimal overhead.

• Docker installed on your machine.

To get started with LLM Playground, you can use Docker to pull and run the container:

docker pull ghcr.io/almogbaku/llm-playground
docker run -p 8080:8080 ghcr.io/almogbaku/llm-playground

This will start the LLM Playground on port 8080.

LLM Playground allows various configuration methods including environment variables, a .env file, or a settings.yml file.

• openai_api_key: Your OpenAI API key.
• openai_organization: Your OpenAI organization ID.
• openai_base_url: Base URL for the OpenAI API.
• models: Configuration for the models and endpoints.

Configure your models using one of the following methods:

1. Direct Configuration: Specify models directly in the models.models parameter.
2. API Provider URLs: Set models.urls to fetch models from an LLM-Playground compatible API (GET request that returns an array of Model).
3. OpenAI API URLs: Set models.oai_urls to fetch models from an OpenAI compatible API.

Each model can be configured with a base_url if it does not utilize OpenAI or is not fetched from models.oai_urls.

Here is a more detailed example using an environment variable setup:

export OPENAI_API_KEY="your-openai-api-key"

export MODELS_MODELS_0_NAME="LLama3"
export MODELS_MODELS_0_DESCRIPTION="Facebook's Llama3 Model"
export MODELS_MODELS_0_TYPE="chat"
export MODELS_MODELS_0_MAX_TOKENS="32000"
export MODELS_MODELS_0_VENDOR="Facebook"

For multiple models, repeat the pattern adjusting the MODEL_#_ prefix.

openai_api_key: "your-openai-api-key"
models:
  urls: ["http://example.com/api/models"] # Fetch models from an LLM-Playground compatible API
  oai_urls: ["http://example.com/api/openai-models"] # Fetch models from an OpenAI compatible API
  models:
    - name: "llama3"
      description: "Facebook's Llama3 Model"
      type: "chat"
      base_url: "https://api.example.com"
      max_tokens: 32000
      vendor: "Facebook"

Once deployed, access LLM Playground by visiting http://localhost:8080. Choose from the available models to start your experiments and comparisons.