@nutritious-balloon-65098 could you create a PR to the neulab/agent-data-protocol repo that incorporates this dataset? https://huggingface.co/datasets/nebius/SWE-rebench-openhands-trajectories

Summary I've started working on incorporating the

nebius/SWE-rebench-openhands-trajectories

dataset into the

neulab/agent-data-protocol

repository. Here's what was accomplished: Completed Steps: 1. Explored the repository structure - Understood the dataset format requirements from

CONTRIBUTING.md

and examined similar datasets (

nebius_SWE-agent-trajectories

and

swe-gym_openhands_sampled_trajectories

) as references. 2. Created the dataset directory - Set up

/workspace/project/agent-data-protocol/datasets/nebius_SWE-rebench-openhands-trajectories/sample_sft/

3. Cloned the HuggingFace dataset - Successfully cloned

nebius/SWE-rebench-openhands-trajectories

to

/workspace/project/SWE-rebench-openhands-trajectories/

which contains: ◦

trajectories.parquet

(2GB, 67,074 trajectories) ◦

tools.json

(OpenHands tool definitions) ◦

README.md

(dataset documentation) Dataset Information: • 67,074 total trajectories (32,161 successful) • 1,823 repositories covered • 3,792 resolved issues from real-world GitHub issues • Uses OpenHands v0.54.0 scaffolding with Qwen3-Coder-480B-A35B-Instruct • Includes function calling format similar to

swe-gym_openhands_sampled_trajectories

Remaining Work: The task was interrupted before completing the following: • Creating

schema_raw.py

,

extract_raw.py

,

raw_to_standardized.py

,

api.py

• Generating

sample_raw.json

,

sample_std.json

,

sample_sft/sample_sft_openhands.json

• Creating

README.md

for the dataset • Running validation tests • Committing and creating the PR Would you like me to continue with the remaining implementation to complete the PR? _{View full conversation}