Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
47 lines (30 loc) · 3.55 KB

CONTRIBUTING.md

File metadata and controls

47 lines (30 loc) · 3.55 KB

Contributing

Microsoft Policy

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repositories using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Contributing to RepoLaunch Source Codes

  1. Welcome issues and PRs related to the bugs and inefficiencies of out agent.

  2. Add unit tests to ./tests folder. Add GitHub workflow to run ./tests automatically.

  3. We found that many threads created from launch/run.py would have "Result Empty Error", which means the last agent state is not saved to disk and not passed back to the main function in launch/run.py. We think it's mostly because docker commit in save.py takes too long time (usually 10min - 120 min) -- it will return read timeout and so often make the thread DEAD... Future works would make docker commit detached in a separate thread/process to solve the problem. Maybe there's also problem in docker concurrency and old Langchain agent apis... Please help us find that problem and fix it!

  4. In launch/utilities/language_handlers.py, you can see language-specific and operating-system-specific prompts and base images. Please help us improve these prompts and add new base images. Base images need update when the latest version of a language updates. Please add official new images if official sources provide them; otherwise you could help us build customized ones and upload to dockerhub public repos, there are example dockerfiles in launch/utilities/dockerfiles.

  5. To improve the success rate / lower down early submit hallucination (unsuccessful build but submit) in the setup stage; and increase the extraction coverage of per-testcase status and per-testcase command from test log in the organize stage -- any suggestions and improvements to the agent workflow is welcome.

Future Directions to Study

We encourage integrating more useful tool calls into RepoLaunch. For example, RAG tools to construct and retrieve memory database of repo launch experiences.

  • The agentic workflow is defined in launch/core/workflow.py
  • The tool calls of each stage are defined in each stage definition file in launch/agent/...
  • We have implemented the string replace editor tool in launch/utilities/tools/str_replace_editor.py. You can add it to the setup agent launch/agent/setup/setup.py if you think for your task fixing the repo bugs during build is necessary. We have not added it into the setup agent because in our task to create SWE tasks, the existing bugs at a buggy commit should be kept as it is, so RepoLaunch should not fix any bugs itself.

We encourage training projects based on the rollout trajectories of RepoLaunch. For example, Rejection Fine-tuning and Reinforcement Learning of open source LMs.

  • The llm calling is defined in launch/utilities/llm.py