jucamohedano/oxford-pets-grpo
Oxford-IIIT Pet — GRPO training data GRPO (verl) training data for Oxford-IIIT Pet breed classification. The model classifies the pet directly; the reward scores a whole-token match of the predicted label plus answer format. Splits: train 2,944 rows, test 3,669 rows. Schema column type data_source string prompt list<element: struct<content: string, role: string>> ability string reward_model struct<ground_truth: string, style: string> images… See the full description on the dataset page: https://huggingface.co/datasets/jucamohedano/oxford-pets-grpo.
Oxford-IIIT Pet — GRPO training data
GRPO (verl) training data for Oxford-IIIT Pet breed classification. The model classifies the pet directly; the reward scores a whole-token match of the predicted label plus answer format.
Splits: train 2,944 rows, test 3,669 rows.
Schema
Format
Parquet in the verl RL schema, one row per image: a chat prompt (system + user), an images column (PNG-encoded), reward_model = {style: rule, ground_truth: <breed>}, ability: classification, and extra_info. Consumed directly by the verl GRPO trainer with data.image_key=images.
Provenance
Built with generate_grpo_dataset() for GRPO training experiments on open-world image classification with large multimodal models (Test-Time Warm-Up line). Derived from the Oxford-IIIT Pet Dataset (Parkhi et al., CVPR 2012).