diffusiongemma
diffusiongemma-opd-prefixes
OPD prefixes
Prompt-only subsets of NVIDIA Nemotron-Post-Training-Dataset-v1 and NVIDIA SWE-Zero-openhands-trajectories, plus codeparrot/APPS and GSM8K training questions. Nemotron tool-call entries preserve native tools in metadata and tokenize through the model native tool template. These are first-decision contexts, not a complete tool interaction environment. Each source retains its own license, not a choice of dual licenses: NVIDIA sources declare CC-BY-4.0; APPS and GSM8K… See the full description on the dataset page: https://huggingface.co/datasets/hydral8/diffusiongemma-opd-prefixes.diffusion-gemma-math-rollout