The video explores the utilization of local models, specifically Llama 3 and Phi-3, for function calling without relying on cloud services. It demonstrates setting up the Llama 3 model with Ollama, using LangChain expression language for structured outputs, and the advantages of running agents locally, such as cost efficiency and avoiding token limits. The video also examines the Phi-3 model’s capabilities in structured output and function calling, comparing it to Llama 3.
- Introduction to local models
- Discusses moving away from cloud services
- Focuses on using Llama 3 and Phi-3 locally
- Highlights the benefits of local model use
- Setting up Llama 3 with Ollama
- Explains the setup process and necessary components
- Demonstrates generating an article using a prompt template
- Shows real-time streaming of model output
- Structured outputs with JSON
- Details the process of defining a JSON schema
- Uses JSON output parser for structured responses
- Compares JSON and string output parsers
- Function calling with Ollama functions
- Introduces tool and function calling with local models
- Utilizes Pydantic for structured outputs
- Compares the performance of Llama 3 and Phi-3 in function calling
Berkeley Function Calling Leaderboard (aka Berkeley Tool Calling Leaderboard)