Add remaining files (#43)

README.md

@@ -1,17 +1,35 @@
-# Magic Commit! ✨
+# Magic Commit! ✨ 🍰
 
-`magic-commit` is a command-line tool for writing your commit messages. It pings OpenAI's GPT-3 API to generate commit messages based on your commit history.
+<img src="splash-image.png" alt="Magic Commit" width="400"/>
+
+`magic-commit` writes your commit messages with AI. 
+
+It's available as a command-line tool currently. There is an experimental VSCode extension in alpha, which you can read about in `Experiments > VSCode Extension` below.
+
+## Table of Contents
+
+- [Installation](#installation)
+- [Setup](#setup)
+- [Usage](#usage)
+- [Experiments](#experiments)
+    - [VSCode Extension](#vscode-extension)
+    - [Llama2 Model](#llama2-model)
+- [Developer Notes](#developer-notes)
+    - [Building the Command-Line Tool](#building-the-command-line-tool)
+    - [Building the VSCode Extension](#building-the-vscode-extension)
+    - [Publishing to PyPI](#publishing-to-pypi)
+    - [Unit Tests](#unit-tests)
 
 ## Installation
 
-All platforms via [PyPI](https://pypi.org/project/magic-commit/)
+To install the command-line tool, [PyPI](https://pypi.org/project/magic-commit/) is the easiest way:
 ```bash
 pip install magic-commit
 ```
 
 ## Setup
 
-You'll need to set up an OpenAI account and get an API key. You can do that on [OpenAI's website](https://platform.openai.com/account/api-keys).
+You'll need to set up an OpenAI account and [get an API key](https://platform.openai.com/account/api-keys).
 
 Once you have a key, add it to `magic-commit` like so:
 ```bash
@@ -22,44 +40,114 @@ magic-commit -k <your-key-here>
 
 Running `magic-commit` is straightforward:
 ```bash
-magic-commit # will run in your current directory
->>> [your commit message] # automatically copied to your clipboard
+>>> magic-commit # will run in your current directory
+[your commit message] # automatically copied to your clipboard
 ```
 
-To specify a directory:
+To see all the options, run:
 ```bash
-magic-commit -d <path-to-git-repo>
+>>> magic-commit --help
+
+usage: magic-commit [-h] [-d DIRECTORY] [-m MODEL] [-k API_KEY] [--set-model MODEL_NAME] [--no-copy] [--no-load] [-t TICKET] [-s START] [--llama LLAMA]
+
+Generate commit messages with OpenAI’s GPT.
+
+optional arguments:
+  -h, --help            show this help message and exit
+  -d DIRECTORY, --directory DIRECTORY
+                        Specify the git repository directory
+  -m MODEL, --model MODEL
+                        Specify the OpenAI GPT model
+  -k API_KEY, --key API_KEY
+                        Set your OpenAI API key
+  --set-model MODEL_NAME
+                        Set the default OpenAI GPT model
+  --no-copy             Do not copy the commit message to the clipboard
+  --no-load             Do not show loading message
+  -t TICKET, --ticket TICKET
+                        Request that the provided GitHub issue be linked in the commit message
+  -s START, --start START
+                        Provide the start of the commit message
+  --llama LLAMA         Pass a localhost Llama2 server as a replacement for OpenAI API
 ```
 
-For help:
-```bash
-magic-commit -h # or --help
-```
+For models, note that:
+- You need to specify an [OpenAI GPT model](https://platform.openai.com/docs/models).
+    - e.g. `gpt-3.5-turbo-0301`, or `gpt-4`
+    - There is an experimental mode which uses Meta's Llama2 models instead. 
+        - (see `Experiments > Llama2 Model` below)
+- Your OpenAI account needs to have access to the model you specify.
+    - i.e. Don't specify `gpt-4` if you don't have access to it.
 
-## Specifying a model
+## Experiments
 
-To change the model for this run:
+### VSCode Extension
+
+Currently in "alpha" status (v 0.0.3). It works, completely, but we need to address the following:
+
+- [ ] Write automated tests
+- [ ] Fix any [known bugs](https://github.com/heyodai/magic-commit/issues?q=is%3Aissue+is%3Aopen+label%3Abug)
+- [ ] Write documentation
+- [ ] Officially publish to the VSCode Marketplace
+
+### Llama2 Model
+
+Llama2 is a free alternative to OpenAI's `GPT-3.5`, created by Meta (Facebook). A long-term goal of `magic-commit` is to support Llama2 fully, allowing you to use it without needing to pay OpenAI or send any potentially sensitive data to them.
+
+To that end, you can pass a running `localhost` Llama2 server to `magic-commit` like so:
 ```bash
-magic-commit -m <model-name> 
+magic-commit --llama http://localhost:8080 # or whatever port you're using
 ```
 
-To change the model globally:
+Note that you'll need to have a running Llama2 server. If you're on MacOS, I found [these instructions](https://github.com/abetlen/llama-cpp-python/blob/main/docs/install/macos.md) from the `llama-cpp-python` project fairly easy to follow.
+
+In the future, the end goal is to seamlessly support both OpenAI and Llama2 models, and to allow you to switch between them with a simple flag.
+
+### LoRA Fine-Tuned Model
+
+Llama2 models capable of running on a normal computer have to be fairly small, e.g. 7 billion parameters. This is a lot, but it's a far cry from the 175 billion parameters of OpenAI's `GPT-3.5` model. Performance for this task out-of-the-box is not great.
+
+However, there is hope. Low-Rank Adaptation (LoRA) is a technique for specializing a large model to a smaller one. To quote the [research paper](https://arxiv.org/abs/2106.09685):
+
+> Compared to GPT-3 175B fine-tuned with Adam, LoRA can reduce the number of trainable parameters by 10,000 times and the GPU memory requirement by 3 times. LoRA performs on-par or better than fine-tuning in model quality on RoBERTa, DeBERTa, GPT-2, and GPT-3
+
+I do believe that we can potentially get GPT-3.5 level of quality while running on a laptop. You can see my experiments with this in the `lora-experiments` folder. If you have any ideas or suggestions, please reach out!
+
+## Developer Notes
+
+Please feel free to open a GitHub issue, submit a pull request, or to reach out if you have any questions or suggestions!
+
+### Building the Command-Line Tool
+
+Note: This is referring to a local development build. For production, see `Publishing to PyPI` below.
+
 ```bash
-magic-commit --set_model <model-name>
+cd cli/magic_commit
+pip install -e . # install the package in editable mode
 ```
 
-For models, note that:
-- You need to specify an [OpenAI GPT model](https://platform.openai.com/docs/models).
-    - e.g. `gpt-3.5-turbo-0301`, or `gpt-4`
-- Your account needs to have access to the model you specify.
+### Building the VSCode Extension
 
-## Contributing
+```bash
+cd vscode/magic-commit
+npm install vsce # if you don't have it already
+vsce package # creates a .vsix file
+```
 
-Feel free to open an issue or submit a pull request.
+### Publishing to PyPI
 
 To publish a new version to PyPI:
 ```bash
-pip install twine wheel # if you don't have it already
+cd cli/magic_commit
+pip install twine wheel
 python setup.py sdist bdist_wheel # build the package
 twine upload dist/* # upload to PyPI
 ```
+
+### Unit Tests
+
+To run the unit tests:
+```bash
+cd cli/magic_commit/tests
+pytest
+```

cli/magic_commit/__main__.py

@@ -40,16 +40,17 @@ def main() -> None:
         action="store_true",
         help="Do not copy the commit message to the clipboard",
     )
+    parser.add_argument(
+        "--no-load", action="store_true", help="Do not show loading message"
+    )
     parser.add_argument(
         "-t",
         "--ticket",
         help="Request that the provided GitHub issue be linked in the commit message",
         type=int,
     )
     parser.add_argument("-s", "--start", help="Provide the start of the commit message")
-    parser.add_argument(
-        "--no-load", action="store_true", help="Do not show loading message"
-    )
+    parser.add_argument("--llama", help="Pass a localhost Llama2 server as a replacement for OpenAI API")
 
     # Decide what to do based on the arguments
     args = parser.parse_args()
@@ -86,6 +87,7 @@ def main() -> None:
             ticket=ticket,
             start=start,
             show_loading_message=show_loading_message,
+            llama2_url=args.llama,
         )
 
         print(results)

cli/magic_commit/magic_commit.py

@@ -5,6 +5,7 @@
 import itertools
 import time
 import sys
+import requests
 
 import openai
 from jinja2 import Environment, PackageLoader
@@ -21,9 +22,14 @@ class GitRepositoryError(Exception):
 class OpenAIKeyError(Exception):
     """Custom exception for OpenAI API key errors."""
 
+
+
+class Llama2ServerError(Exception):
+    """Custom exception for Llama2 server errors."""
     pass
 
 
+
 def is_git_repository(directory: str) -> bool:
     """
     Check if a directory is a Git repository.
@@ -135,7 +141,7 @@ def get_commit_messages(directory: str) -> str:
 
 
 def generate_commit_message(
-    diff: str, start: str, ticket: str, api_key: str, model: str
+    diff: str, start: str, ticket: str, api_key: str, model: str, llama2_url: str = None
 ) -> str:
     """
     Generate a commit message.
@@ -152,6 +158,8 @@ def generate_commit_message(
         The OpenAI API key.
     model : str
         The OpenAI GPT model to use.
+    llama2_url: str
+        (Optional) The URL of the Llama2 server to use.
 
     Returns
     -------
@@ -176,9 +184,18 @@ def generate_commit_message(
         {"role": "system", "content": system_msg},
         {"role": "user", "content": user_msg},
     ]
-    openai.api_key = api_key
-    response = openai.ChatCompletion.create(model=model, messages=messages)
-    response = response.choices[0].message.content.strip()
+
+    # Determine which service to use
+    if llama2_url:
+        # Call the Llama2 server
+        response = call_llama2_server(llama2_url, messages)
+        print(response)
+        response = response['choices'][0]['message']['content'].strip()
+    else:
+        # Use OpenAI's service
+        openai.api_key = api_key
+        response = openai.ChatCompletion.create(model=model, messages=messages)
+        response = response.choices[0].message.content.strip()
 
     # Strip the first line of response
     # Assign it to start if it is empty
@@ -193,6 +210,36 @@ def generate_commit_message(
     return render_final_template(start, response, ticket).strip()
 
 
+
+def call_llama2_server(url: str, messages: list) -> dict:
+    """
+    Call the Llama2 server.
+
+    Parameters
+    ----------
+    url : str
+        The URL of the Llama2 server.
+    messages : list
+        The messages to send to the server.
+
+    Returns
+    -------
+    dict
+        The response from the server.
+
+    Raises
+    ------
+    Llama2ServerError
+        If an error occurs while connecting to the server.
+    """
+    try:
+        response = requests.post(url, json={"messages": messages})
+        response.raise_for_status()
+        return response.json()
+    except requests.exceptions.RequestException as e:
+        raise Llama2ServerError(f"An error occurred while connecting to the Llama2 server: {e}")
+
+
 def render_template(message: str, template_name: str) -> str:
     """
     Render the commit message template.
@@ -275,6 +322,7 @@ def run_magic_commit(
     api_key: str,
     model: str,
     show_loading_message: bool,
+    llama2_url: str = None
 ) -> str:
     """
     Generate a commit message and return it.
@@ -293,6 +341,8 @@ def run_magic_commit(
         The OpenAI GPT model to use.
     show_loading_message : bool
         Whether or not to show the loading animation.
+    llama2_url: str
+        (Optional) The URL of the Llama2 server to use.
 
     Returns
     -------
@@ -309,7 +359,7 @@ def run_magic_commit(
         diff = run_git_diff(directory)
         if not check_git_status(directory):  # Check if there are staged changes
             return "⛔ Warning: No staged changes detected. Please stage some changes before running magic-commit."
-        commit_message = generate_commit_message(diff, start, ticket, api_key, model)
+        commit_message = generate_commit_message(diff, start, ticket, api_key, model, llama2_url)
     finally:
         # Ensure the loading animation stops
         if show_loading_message:

cli/setup.py

@@ -6,7 +6,7 @@
 
 setup(
     name="magic-commit",
-    version="0.5.1",
+    version="0.6.1",
     packages=find_packages(),
     include_package_data=True,  # This line is needed to include non-code files
     package_data={