LangChain Tool Calling feature just changed everything

---

---

AI Summary

• Overview of Lang Jan’s Tool Calling Feature
  • Underrated but offers significant flexibility.
  • Allows switching between models supporting function calling.
• Previous Limitations
  • Function calling was tailored to OpenAI’s API.
  • Other vendors like Vertex, Gemini, and Anthropic Sonet had different schemas.
  • Integration with non-OpenAI models was troublesome.
• Lang Jan’s Solution
  • Created a unified interface for tool calling.
  • Supports multiple models: OpenAI, Vertex, Gemini, Mistral, Fireworks, Anthropic Sonet.
  • Responds to community requests for vendor flexibility.
• Interface Components
  • bind function method: Allows LLM to use custom functions.
  • tool calls: Populated with function call invocations by the LLM.
  • Tool Calling Agent: Creates a function calling agent for any supported vendor.
• Example Usage
  • Define tools or use OpenAI format for function calling.
  • Bind tools to the LLM.
  • LLM decides on invoking functions during calls.
• Agent Creation
  • Function calling agent can be created with different models like Anthropic Sonet.
• Demo Explanation
  • Defined tools: multiply and Tab Search for real-time data.
  • Initialized tool calling agent with OpenAI’s GPT-4 and Anthropic Sonet.
  • Asked for weather comparison in Dubai and San Francisco in Celsius.
  • Results traced with LSmith for examination.
• Results
  • OpenAI and Anthropic Sonet provided similar weather results.
  • Traces showed the process of invoking tools and summarizing results.
• Conclusion
  • Lang Jan’s implementation simplifies switching between models with function calling.
  • It’s a significant advancement in machine learning commoditization.
  • Fulfills community demand for working with various models.