datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
EgoLoc-Contact-GRPO
EgoLoc Contact Exact-Moment Grid GRPO
This is a self-contained 3x3 image-grid dataset for GRPO training on exact
contact/start localization. The numbered cells are chronological and use 1-based
indices.
This dataset is used to improve a VLM's accuracy for the EgoLoc pipeline.
This dataset IS NOT shuffled. When undergoing GRPO, recommend shuffling the dataset.
3x3 grid dataset for VLM tuning on contact frame identification.
Splits
Training rows: 1389
Validation… See the full description on the dataset page: https://huggingface.co/datasets/yuchenxie/EgoLoc-Contact-GRPO.surfer-grpo
Kedar84/surfer-grpo
Source run file: dataset-run-1758885513916.jsonl
Generated: 2025-09-27 (UTC)
Fields:
url: Source page URL gathered by Surfer automation.
bounding_box: Normalized viewport coordinates for the target element.
raw_ss: PNG screenshot of the raw page stored in images/raw/.
annotated_ss: PNG screenshot with bounding boxes overlayed (images/annotated/).
Loading Example
from datasets import load_dataset
ds = load_dataset("Kedar84/surfer-grpo"… See the full description on the dataset page: https://huggingface.co/datasets/Kedar84/surfer-grpo.simplevla-grpo-assets
SimpleVLA GRPO Grasp Assets
This dataset contains the released USD object assets used by the SimpleVLA-style GRPO grasping experiments.
Expected local layout after running scripts/download_assets.sh:
/data4/nerako/reasoning/RLinf_assets/grasp_assets/
medical-rl-grpo-v1VHM_dataset_grpoEgoLoc-Separation-GRPO
EgoLoc Separation GRPO Dataset
This is a self-contained 3x3 image-grid dataset for GRPO training on exact
separation/end localization. The numbered cells are chronological and use 1-based
indices.
This dataset is used to improve a VLM's accuracy for the EgoLoc pipeline.
This dataset IS NOT shuffled. When undergoing GRPO, recommend shuffling the dataset.
3x3 grid dataset for VLM tuning on separation frame identification.
Splits
Training rows: 1127
Validation rows:… See the full description on the dataset page: https://huggingface.co/datasets/yuchenxie/EgoLoc-Separation-GRPO.ETCHR-GRPO-10K
ETCHR GRPO-10K
📖Paper
| 🏠Homepage
| 🤗ETCHR-FLUX.2-klein-9B Model
| 🤗ETCHR SFT-400K Dataset
| 🤗ETCHR GRPO-10K Dataset
| 🤗DL3DV-2K Benchmark
ETCHR GRPO-10K is the GRPO training data for further enhance ETCHR's editing capaibility in assisting understanding models. It contains 10000 samples of five tasks (Fine-grained Perception, Chart Understanding, Maze Solving, Jigsaw Puzzle and Spatial Understanding). Each sample contains the image to be edited, an editing… See the full description on the dataset page: https://huggingface.co/datasets/internlm/ETCHR-GRPO-10K.chart-rvr-grpo-train
Chart Reasoning Dataset
The dataset has been curated to train Chart-RVR-Hard models using GRPO configuration.
The training data is strictly from ChartQA, PlotQA and ChartFC.
membrane-grpo-results
Membrane — GRPO results & eval
Browse files: showcase/ (all SVGs below are the same paths).
The Dataset Viewer tab is off here: that UI expects a single table or raster image split, while this repo is mostly checkpoints and CSVs under runs/ / eval/. Previews on this card use normal Markdown images instead.
Artifacts: eval/, runs/, existing_run/.
Space: Tejasghatule/membrane-temp
All showcase/ figures
reviewer_results_overview.svg
One page: eval, Colab hero… See the full description on the dataset page: https://huggingface.co/datasets/Tejasghatule/membrane-grpo-results.ShowUI-web-8k-grounding-grpollavaone_grpo_v1vlaa-thinking-grpo
VLAA-Thinking-SFT-126K
Large-scale vision-language dataset with 126K instruction-following samples featuring chain-of-thought reasoning
Dataset Description
This dataset contains vision-language samples with instruction-following conversations. Each sample includes:
image: PIL Image object
question: Question or instruction text
answer or gt: Response with thinking process (SFT dataset) or ground truth answer (GRPO dataset)
caption: Image caption (may be empty for some… See the full description on the dataset page: https://huggingface.co/datasets/penfever/vlaa-thinking-grpo.sceneteract-grpo
SceneTeract GRPO Training Set
Action-level feasibility samples for post-training a VLM against a geometric verifier.
Each row is one atomic interaction — an image, a prompt, and a label that was measured
rather than annotated — ready to drop into TRL's GRPOTrainer.
8,073 samples over 1,132 3D-FRONT living rooms and dining rooms and three agent profiles.
from datasets import load_dataset
ds = load_dataset("leopoldmaillard/sceneteract-grpo")
ds["train"] # 6,473 samples / 905… See the full description on the dataset page: https://huggingface.co/datasets/leopoldmaillard/sceneteract-grpo.llavaone_grpo_v2Flow-GRPOshotpath-grpo-trajectory-audit-20260729
ShotPath GRPO trajectory audit
This audit reconstructs the committed 260-step trajectory by keeping the last logged occurrence of each pair after time-limit rollbacks. It includes compact statistics for every committed group and detailed candidate/judge records plus pre/post images for stratified and contrast samples.
LLaVA-Critic-GRPO-datasetoxford-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.NFA_OCR_qwen_grpo_formatv2jee-grpo-v1
JEE-GRPO-v1: Visual Reasoning Dataset for VLMs
Dataset Summary
JEE-GRPO is a high-quality, multimodal dataset designed for training and benchmarking Vision Language Models (VLMs) on complex STEM problems.
Unlike traditional text-only datasets, this dataset renders JEE Main (Joint Entrance Examination) questions as high-resolution images. This approach bypasses OCR errors and perfectly preserves complex LaTeX equations, diagrams, and chemical structures, making it ideal… See the full description on the dataset page: https://huggingface.co/datasets/farhananis005/jee-grpo-v1.plantnet300k_grpooxford-pets-grpo-think
Oxford-IIIT Pet — GRPO training data (structured reasoning)
GRPO (verl) training data for Oxford-IIIT Pet breed classification with a structured-reasoning prompt: the model emits a scratchpad tagging visible properties (HasProperty), parts (HasA), and setting (AtLocation) before the label. Reward: 0.30 for a well-formed think block, 0.70 for the label match.
Splits: train 2,944 rows, test 3,669 rows.
Schema
column
type
data_source
string
prompt… See the full description on the dataset page: https://huggingface.co/datasets/jucamohedano/oxford-pets-grpo-think.OpenVLThinker-grpo-hardMath-Llava-GRPO-random-50k-augmentationOpenVLThinker-grpo-mediumNFA_OCR_qwen_grpo_format1grpo-single-binary-1kUiPad-GRPOGRPODatasetGRPO_val_benchmark