modelco 0.0.1.5

LLM collaboration

model_collaboration

The future is now.

Contributors who just joined us, read this

Thank you for joining this effort! Kindly do the following:

1. Get familiar with "Github Setup" in this readme: you need to create your own branch, code in your branch, and submit pull requests from your branch to the dev branch. Never edit anything on the dev branch.
2. Follow "After that, quick start!!" in this readme.
3. Read the beginning of model_collaboration/method/user_readme.md and briefly check out the methods we already have here.
4. Follow model_collaboration/method/sample_approach.py and model_collaboration/method/developer_readme.md to develop your own method. Try to use the helper functions provided there if possible (such as distributed_generation.distributed_generation).
5. After you implemented and tested your method, git add commit push, open a pull request from your branch to dev, assign your point of contact and Shangbin as the two reviewers.

Github Setup

1. Clone the repo with https://github.com/BunsenFeng/model_collaboration.git.
2. Checkout to the dev branch: git checkout dev
3. Pull the latest changes from the dev branch to your local dev branch: git pull
4. Create your own feature/hotfix branch on local: git checkout -b [your-local-branch-name]
5. Make edits on the scripts you care.
6. Push any changes you made on your local branch to the GitHub server - after git add and git commit operations, do git push, you will see git push --set-upstream origin [your-local-branch-name] suggested by github, copy and paste this command and run.
7. Open a new Pull Request from the GitHub webpage, make sure it's merging from [your-local-branch-name] to the dev branch. Add any reviewer and Shangbin that matters to the changes.
8. Once approved, merge the changes to the dev branch.
9. After merging, you will see an option on the webpage to delete your own branch. Delete it.
10. Loop from #2.

If you are in the middle of the development, and you need the latest changes from dev branch, follow the steps below:

1. Keep track of the current changes you made on your local branch: git add and git commit your [your-local-branch-name]
2. Checkout to the dev branch: git checkout dev
3. Pull the latest changes from the dev branch: git pull
4. Check back to your local branch: git checkout [your-local-branch-name]
5. Merge the changes from dev branch to your own branch: git merge dev
6. Keep working on your own branch. done.

Questions about git? Don't take guesses, email [e-mail] and cc [e-mail].

After that, quick start!!

conda env create -f environment.yml
conda activate model_collaboration
cd ..
git clone https://github.com/arcee-ai/mergekit.git
cd mergekit
pip install -e .
cd ..
cd model_collaboration

Run your first model collaboration experiment (if you don't have 3 GPUs, go to test_config.json and set "gpu_ids": [0], [0,1], or whatever you have; if your GPU is nice, increase batch_size):

python -m model_collaboration.main -c model_collaboration/test_config.json

You will see the outputs and evaluation results in the model_collaboration/logs/ folder.

See model_collaboration/method/user_readme.md for more details about different collaboration methods implemented.

Zhaoxuan (our evaluation tsar), additionally see model_collaboration/data/eval_readme.md.