OfficeWorld dataset (needed for SFT+RL pipelines)

[vyomakesh0728]

hey @hanyullai and @sdpkjc ,

I'm currently contributing to environments hub as part of Prime-Intellects community, and would love to integrate ComputerRL into verifiers toolkit for agent benchmarking. We've hit two blockers:

1. No license file: Without an explicit open source license, we can't reuse or extend ComputerRL, even though it builds on OSWorld (Apache-2.0). Could you clarify or add a license?

2. Packaging improvement: I've drafted a minimal pyproject.toml to support modern python/uv-based installs and submitted a PR for review.

Adding a license and supporting pyproject.toml would enable much wider research adoption and reproducibility.

Could you please look into this issue and help us incorporate ComputerRL?

[hanyullai]

Hi @vyomakesh0728,

Thanks a lot for your feedback and contributions!
We’ve added an explicit open source license to ComputerRL and merged your pyproject.toml improvements to support modern packaging.
Both changes are now available in the latest version on the main branch.

Really appreciate your efforts to help bring ComputerRL into the wider research community — looking forward to seeing it integrated into the verifiers toolkit!

[vyomakesh0728]

Hi @vyomakesh0728,

Thanks a lot for your feedback and contributions! We’ve added an explicit open source license to ComputerRL and merged your pyproject.toml improvements to support modern packaging. Both changes are now available in the latest version on the main branch.

Really appreciate your efforts to help bring ComputerRL into the wider research community — looking forward to seeing it integrated into the verifiers toolkit!

Awesome, thanks for moving on this! we'll get ComputerRL plugged into Environments Hub with verifiers toolkit soon and update here if we run into anything.

[vyomakesh0728]

hey @hanyullai again,

Appreciate the super quick turnaround on license/toml! One more quick thing: Would it be possible to release the full OfficeWorld dataset (needed for SFT+RL pipelines)?

Its kinda required for us to train with prime-rl infra to fully integrate ComputerRL into the Environments Hub with verifiers toolkit.

Thanks in advance for taking the time and supporting us.