| --- |
| license: apache-2.0 |
| task_categories: |
| - text-generation |
| tags: |
| - agent |
| - harbor |
| - reinforcement-learning |
| - nemotron |
| --- |
| |
| # laion/nemotron-gym-arc-agi-python-inductive |
|
|
| Harbor task-binary dataset (10,000 tasks) converted from **nvidia/Nemotron-RL-ARC-AGI-v1 [python_inductive]** |
| (part of the [nvidia/Nemotron-Post-Training-v3](https://huggingface.co/collections/nvidia/nemotron-post-training-v3) collection). |
| |
| Each row is a valid [Harbor](https://github.com/open-thoughts/OpenThoughts-Agent) |
| task binary: columns `path` (str) and `task_binary` (gzip tar). Converted with the |
| OpenThoughts-Agent `data.nemotron_gym` framework. |
| |
| **Grading:** Sandboxed exec of agent transform() vs held test grids. |
| |
