Storia-AI/sage

Feature request: Linkify code symbols in the model response

Aperta

#54 aperta il 4 ott 2024

 (0 commenti) (0 reazioni) (0 assegnatari)Python (116 fork)auto 404
hacktoberfest🔥 difficulty: 3

Metriche repository

Star
 (1267 stelle)
Metriche merge PR
 (Metriche PR in attesa)

Descrizione

Given an LLM response that contains a code symbol (e.g. class name, method, etc.), we should link it to its exact location on GitHub. For instance, if you're chatting with Hugging Face's Transformers library:

  • Original response: "To define a BERT model using the Hugging Face Transformers library, you can use the BertModel class."

  • Link-ified response: "To define a BERT model using the Hugging Face Transformers library, you can use the BertModel class."

This should be done as a post-processing step in chat.py, once the model is done streaming.

Work items:

  1. Produce an AST (Abstract Syntax Tree) for every file in the repository. This can be done via the tree_sitter library, which we're already using in chunk.py.
  2. Identify class / method / file names in the model response, potentially using regular expressions.
  3. Look up the strings matched by the regular expressions in the per-file ASTs.
  4. Map the code symbol to a github URL.

Guida contributor