Issue Description

When switching between different embedding models (e.g., from OpenAI to Ollama), users are encountering dimension mismatch errors in ChromaDB. This occurs because ChromaDB gets initialized with a fixed dimension (1536 for OpenAI) and doesn't handle different embedding dimensions gracefully.

Current Behavior

• ChromaDB gets locked to the dimension of the first embedding model used (typically 1536 for OpenAI)
• Switching to a different model with different dimensions causes errors
• No clear warning or handling mechanism for dimension mismatches

Proposed Solution

We should:

1. Add a check for embedding dimensions before initializing ChromaDB
2. Provide clear error messages when dimension mismatch occurs
3. Add documentation about handling different embedding models
4. Consider adding an automatic cleanup/reinit mechanism when switching models

Tasks

• Add dimension validation checks
• Implement clear error messages
• Update documentation with model switching guidelines
• Consider adding a utility function to handle DB cleanup

Temporary Workaround

For users encountering this issue:

1. Delete the contents of the data folder to clear ChromaDB
2. Restart training with the new embedding model

Related Issues

#134