Open Interpreter Launches Local III with Enhanced Local Model Support and Open-Source Initiatives

Open Interpreter Launches Local III with Enhanced Local Model Support and Open-Source Initiatives

June 18, 2024

Open Interpreter has just released Local III, a major update that significantly improves the reliability of local models and marks the beginning of their journey towards an open-source language model. This release is a significant step forward in providing developers with personal, private access to machine intelligence.

Technical Changes and Why They Matter

Enhanced Local Model Explorer

• Interactive Setup: Local III introduces an easy-to-use local model explorer that simplifies the process of selecting and managing inference providers and models.
• Model Management: You can now download new models directly from the explorer, making it easier to keep your setup up-to-date.

Why It Matters: For developers, this means a more streamlined workflow. No more manual downloads or complex configuration steps. Just a simple, interactive interface that guides you through the process.

Free Hosted i Model

• Llama3-70B Endpoint: Local III includes a free language model endpoint serving Llama3-70B via the command interpreter --model i.
• Opt-In Training Data: Conversations with the i model will be used to train an open-source language model for computer control. Personally identifiable information (PII) will be removed before open-sourcing.

Why It Matters: This provides a no-setup experience for users while contributing to the development of an open-source AI. It's a win-win for both individual developers and the broader community.

Deep Integration with Ollama

• Unified Command: To give any Ollama model access to a code interpreter, simply run interpreter --model <model>, where <model> is from Ollama’s model library.
• Model Download Management: The command will only download the model if you haven’t already done so.

Why It Matters: This integration simplifies the process of using different models, reducing the overhead for developers and making it easier to experiment with various language models.

Optimized Profiles for SOTA Models

• Recommended Settings: Local III includes optimized profiles for Codestral, Llama3, and Qwen. You can configure Open Interpreter to use these settings with the following commands:
  • interpreter --profile codestral.py
  • interpreter --profile llama3.py
  • interpreter --profile qwen.py

Why It Matters: These profiles ensure that you get the best performance out of your local models without having to manually tune settings. If you find optimal settings for other models, consider contributing a PR to the default profiles folder.

How to Get Started

1. Start the Local Explorer:

   interpreter --local
   

2. Use the i Model:

   interpreter --model i
   

3. Integrate Ollama Models:

   interpreter --model <ollama_model>
   

4. Apply Optimized Profiles:

   interpreter --profile <profile_name>.py
   

Community and Contributions

Local III is the result of a collaborative effort involving over 100 contributors from around the world. The project is open to contributions, and you can help by:

• Making PRs to improve the local explorer.
• Contributing optimal settings for other local language models.

Conclusion

Local III represents a significant leap forward in making local machine intelligence more accessible and reliable. By simplifying model management, providing a free hosted endpoint, and integrating with leading inference engines, Open Interpreter is empowering developers to take control of their AI tools. Whether you're just starting out or are an experienced developer, Local III has something to offer.