Skip to content

256 token limit issue on MiniLM (urgent) #5

Description

@ramsterr

MiniLM can handle like 40 lines of code in one function

so lets a say if a function is bigger , and maybe around 200 lines (assuming that im indexing a big repo) , then the embeddings would be terrible.

so some suggested solutions:

-write a chunker
-Detects functions longer than ~1,200 chars
-Splits them into overlapping 50-line windows (overlap prevents splitting a sentence mid-thought)
-Creates multiple entries for one function — each chunk gets its own vector
-On search, if any chunk matches, you show the full function

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions