Finetuned Model 2

This model has a top-1 accuracy of 93% on the test split and is improved from the previous released model.

Finetuned Model 1

This release contains a model finetuned for agent selection.

This model is significantly better than the base model at resolving edge cases correctly, such as assigning "my teeth have calculus, how do I fix this?" to the health agent instead of the math agent. The foundational model is sentence-transformers/all-mpnet-base-v2.

Place the "test_model" folder in ./models/ inside the repository, and then run the test.py script while in the virtual environment to try this model out.