-
-
Save karpathy/1dd0294ef9567971c1e4348a90d69285 to your computer and use it in GitHub Desktop.
# ----------------------------------------------------------------------------- | |
# AI-powered Git Commit Function | |
# Copy paste this gist into your ~/.bashrc or ~/.zshrc to gain the `gcm` command. It: | |
# 1) gets the current staged changed diff | |
# 2) sends them to an LLM to write the git commit message | |
# 3) allows you to easily accept, edit, regenerate, cancel | |
# But - just read and edit the code however you like | |
# the `llm` CLI util is awesome, can get it here: https://llm.datasette.io/en/stable/ | |
gcm() { | |
# Function to generate commit message | |
generate_commit_message() { | |
git diff --cached | llm " | |
Below is a diff of all staged changes, coming from the command: | |
\`\`\` | |
git diff --cached | |
\`\`\` | |
Please generate a concise, one-line commit message for these changes." | |
} | |
# Function to read user input compatibly with both Bash and Zsh | |
read_input() { | |
if [ -n "$ZSH_VERSION" ]; then | |
echo -n "$1" | |
read -r REPLY | |
else | |
read -p "$1" -r REPLY | |
fi | |
} | |
# Main script | |
echo "Generating AI-powered commit message..." | |
commit_message=$(generate_commit_message) | |
while true; do | |
echo -e "\nProposed commit message:" | |
echo "$commit_message" | |
read_input "Do you want to (a)ccept, (e)dit, (r)egenerate, or (c)ancel? " | |
choice=$REPLY | |
case "$choice" in | |
a|A ) | |
if git commit -m "$commit_message"; then | |
echo "Changes committed successfully!" | |
return 0 | |
else | |
echo "Commit failed. Please check your changes and try again." | |
return 1 | |
fi | |
;; | |
e|E ) | |
read_input "Enter your commit message: " | |
commit_message=$REPLY | |
if [ -n "$commit_message" ] && git commit -m "$commit_message"; then | |
echo "Changes committed successfully with your message!" | |
return 0 | |
else | |
echo "Commit failed. Please check your message and try again." | |
return 1 | |
fi | |
;; | |
r|R ) | |
echo "Regenerating commit message..." | |
commit_message=$(generate_commit_message) | |
;; | |
c|C ) | |
echo "Commit cancelled." | |
return 1 | |
;; | |
* ) | |
echo "Invalid choice. Please try again." | |
;; | |
esac | |
done | |
} |
here is fork with ollama
+ llama3.1:70b
https://gist.github.com/nikolaydubina/12e3c692eeb3a651579c9f6c25d024f8
UPD:
@thfrei you are right. 405b does not work locally even with 128GB RAM. sigh. got to use 70B instead.
here is fork with
ollama
+llama3.1:405b
https://gist.github.com/nikolaydubina/12e3c692eeb3a651579c9f6c25d024f8
Just out of curiosity, how you run that with ollama and llama3.1:405b locally? Must be quite some gpu-cluster, your workstation? :-)
In case anyone runs into the same error as me, you might need to unalias gcm before the gist above:
# Unalias gcm to allow for custom function
unalias gcm 2>/dev/null
For me this was set to git checkout main
previously by the Oh My Zsh plugin
Fork for fish with basic support for git flags: https://gist.github.com/knyazer/675e6eb945ae5ec64af2f9be4826b07e
I've been using this one for a while. it has a little more detail on how a commit message should be formatted.
git commit . -m "$(git diff | sgpt --model gpt-4o --code 'Write concise, informative commit messages: Start with a summary in imperative mood, explain the 'why' behind changes, keep the summary under 50 characters, use bullet points for multiple changes, avoid using the word refactor, instead explain what was done, and reference related issues or tickets. What you write will be passed to git commit -m "[message]"')"
what about small models with 7b parametres, is it will be good at it?
you can also accomplish this using code ~/.gitconfig
and adding:
[alias]
ai = "!f() { git add . && if [ -n \"$(git diff --cached)\" ]; then git commit -m \"$(git diff --cached | llm -m '4o-mini' 'Below is a diff of all staged changes, coming from the command:\\n```\\ngit diff --cached\\n```\\nPlease generate a concise, one-line commit message for these changes.')\" && git push; else echo 'No changes to commit'; fi }; f"
for some reason the zshrc option lost my local branch context but .gitconfig worked
Additionally, you can also set up a vscode keybinding (cmd + shift + P on mac) to open up the JSON:
// Place your key bindings in this file to override the defaultsauto[]
[
{
"key": "cmd+enter",
"command": "workbench.action.terminal.sendSequence",
"args": { "text": "git ai\n" }
},
....
now you can just do "cmd + enter" to get commit messages quickly.
It goes without saying that you should install [llm](https://llm.datasette.io/en/stable/)
locally via brew or another method and connect your openai api key.
I have been using:
git commit -m "$(git diff --staged | sgpt "Write a single conventional commits style commit message for this diff on the branch $(git rev-parse --abbrev-ref HEAD).")"
Extremely useful!
(1) I have replaced llm
with ollama run llama3.1
.
(2) Something was already taking the gcm command, I added unalias gcm 2>/dev/null
before.
check this https://github.com/Fl0p/gitpmoji
generate commit message / decorate with emoji / give rating / evaluate diff
pros - uses git hook don't need any additional. cons - uses only openAI api
So cool
For oh-my-zsh users, on .zshrc
I added
# must unalias all ZSH defaults here AFTER we source the above
unalias gcm
after
source $ZSH/oh-my-zsh.sh
To make sure the gcm
alias stays ripped out between sessions.
Forked this no-brainer cheap solution for Git commits!
Google Gemini works great with it, especially with the free tier. For testing, use gemini-1.5-flash. Here is my fork,
https://gist.github.com/HtutLynn/7fe83d64d6381726ac06e9fb7f972643
Quick start:
Install the Gemini plugin:
llm install llm-gemini
Set up your API key.
Happy committing! π
@karpathy i make the nodejs version https://github.com/koisose/auto-commit-gaia no need for api key coz its using gaianet.ai also using llama
Forked this no-brainer cheap solution for Git commits!
Google Gemini works great with it, especially with the free tier. For testing, use gemini-1.5-flash. Here is my fork,
https://gist.github.com/HtutLynn/7fe83d64d6381726ac06e9fb7f972643
Quick start:
Install the Gemini plugin:
llm install llm-gemini
Set up your API key.
Happy committing! π
@HtutLynn cant agree more i create both for gemini and groq as well the groq one using gemma and its super fast as well https://github.com/koisose/auto-commit-gemini and https://github.com/koisose/auto-commit-groq
guys, use https://github.com/di-sukharev/opencommit, it's the most feature rich open-source tool
run npm i -g opencommit
, then oco config set OCO_OPENAI_API_KEY=your_key
and then sumply run oco
to generate your git commit messages.
πππ
guys, use https://github.com/di-sukharev/opencommit, it's the most feature reach open-source tool
run
npm i -g opencommit
, thenoco config set OCO_OPENAI_API_KEY=your_key
and then sumply runoco
to generate your git commit messages.πππ
Oh, nice job! Thxπ
guys, use https://github.com/di-sukharev/opencommit, it's the most feature reach open-source tool
run
npm i -g opencommit
, thenoco config set OCO_OPENAI_API_KEY=your_key
and then sumply runoco
to generate your git commit messages.πππ
ππ
guys, use https://github.com/di-sukharev/opencommit, it's the most feature reach open-source tool
run
npm i -g opencommit
, thenoco config set OCO_OPENAI_API_KEY=your_key
and then sumply runoco
to generate your git commit messages.πππ
Oh, I was looking for this repo but forgot the name, thanks!
guys, use https://github.com/di-sukharev/opencommit, it's the most feature reach open-source tool
runnpm i -g opencommit
, thenoco config set OCO_OPENAI_API_KEY=your_key
and then sumply runoco
to generate your git commit messages.
πππOh, I was looking for this repo but forgot the name, thanks!
You are welcome π©
check this https://github.com/Fl0p/gitpmoji generate commit message / decorate with emoji / give rating / evaluate diff pros - uses git hook don't need any additional. cons - uses only openAI api
take a look at https://github.com/di-sukharev/opencommit, we even support llama running locally or remotely
I have been using:
git commit -m "$(git diff --staged | sgpt "Write a single conventional commits style commit message for this diff on the branch $(git rev-parse --abbrev-ref HEAD).")"
π
Made some changes so it doesn't make calls when there are no staged changes. Renamed the command to gai
due to conflict with the Oh My ZSH alias
https://gist.github.com/vlameiras/8591bf75990571f5f1e2ea9c452047c3
Similar tool at coder/aicommit that follows a repo's existing commit style.
i suggest adding a parameter to bypass the ai-generated message when the user passes the -m
flag
so that gcm -m 'my user-generated commit msg'
allows a user to enter their own message without pinging the llm.
to do that you can add the following to line 11.
# Check if -m flag is provided
if [ "$1" = "-m" ] && [ -n "$2" ]; then
# Use the provided message directly
if git commit -m "$2"; then
echo "Changes committed successfully with your message!"
return 0
else
echo "Commit failed. Please check your message and try again."
return 1
fi
fi
Here's a more convenient git ai copilot: https://github.com/Undertone0809/gcop
Benefits:
- Easier to use with git alias, which means just use the git command
- Compatible with all LLMs, easy to configure no matter what model you're using.
- Various git alias extensions to help you use git commands more easily.
To use 4o instead of 4o-mini.
Change:
git diff --cached | llm "
to:
git diff --cached | llm -m "4o" "