---
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.