git-semantic-similarity 1.0.0

file: README.md

git-semantic-similarity

Search git commit messages by semantic similarity with sentence-transformers.

Embeddings are stored on disk for faster retrieval, and can easily be checked into git.

$ gitsem "project scaffolding"

Commit 403836d2ee4900579b0d1e8169dd4bfebddab0ba
Author: Adrian Meidell Fiorito <[e-mail]>
Date:   2024-09-23 19:08:05
Similarity: 0.2299

    Change model, add src folder

Commit d2909a8ec352a881ab05cab8b8a67038b063f37a
Author: Adrian Meidell Fiorito <[e-mail]>
Date:   2024-09-23 19:08:05
Similarity: 0.2086

    Initial commit

...

Commit a09923166072aca4910e92272ef161e3398b1d89
Author: Adrian Meidell Fiorito <[e-mail]>
Date:   2024-09-23 19:08:05
Similarity: -0.0716

    Remove buggy rounding

Installation

Clone and run locally

git clone https://github.com/adrianmfi/git-semantic-similarity.git
cd git-semantic-similarity
pip install .

Usage

In a git repository, run: gitsem "query string"

To only show the 10 most relevant commits:

gitsem "changes to project documentation" -n 10

To use another pretrained model, for example a smaller and faster model:

gitsem "user service refactoring" --model sentence-transformers/all-MiniLM-L6-v2

A list of supported models can be found here

The tool supports forwarding arguments to git rev-list For example, to only search in the 10 most recent commits:

gitsem "query string" -- -n 10

Or to filter by a specific author:

gitsem "query string" -- --author bob

Or you can format the output in a single line for further shell processing:

gitsem "query string" --sort False --oneline -- n 100 | sort -n -r | head -n 10

Arguments

• -m, --model [STRING]:
  A sentence-transformers model to use for embeddings. Default is all-mpnet-base-v2.

• -c, --cache [BOOLEAN]:
  Whether to cache commit embeddings on disk for faster retrieval. Default is True.

• --cache-dir [PATH]:
  Directory to store cached embeddings. If not specified, defaults to git_root/.git_semsim/model_name.

• --oneline:
  Use a concise output format.

• --sort [BOOLEAN]:
  Sort results by similarity score. Default is True.

• -n, --max-count [INTEGER]:
  Limit the number of results displayed. If not provided, no limit is applied.

• -b, --batch-size [INTEGER]:
  Batch size for embedding commits. Default is 1000.

• query [STRING]:
  The query string to compare against commit messages.

• git_args [STRING...]:
  Arguments after -- will be forwarded to git rev-list.

License

MIT