Request for Dense Retrieval Implementation Details

[noahdenicola]

Hi team,

Thank you for releasing REPOCOD and providing the inference code example.

I'm interested in reproducing the dense retrieval mechanism that uses text-embedding-3-small. The current example only uses the "current file" context approach, I was hoping to reproduce results with the dense retrieval method as well.

Would it be possible to:

1. Share the implementation of the dense retrieval mechanism?
2. Provide details such as:

• How different parts of functions (signature/docstring/body) are combined for embedding
• The exact pre-processing steps used before embedding
• The similarity search implementation
• Any specific thresholds or parameters used for retrieval

This would be really helpful to us and the community!

Thank you for considering this request!