datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
soft-toy-wbcd-khlptoy_struc_datasettoy-models-of-sft-data
Toy Models of SFT Data
This is a public-clean candidate data package for the Toy Models of SFT project.
It is built for researcher inspection first.
The package answers two questions:
What were the models trained on?
How did the models actually behave under evaluation?
The package includes training data, eval inputs, model rollouts, judge scores,
parsed GPQA outputs, aggregate tables, paper figures, frozen plot data, and
provenance records. It deliberately includes some… See the full description on the dataset page: https://huggingface.co/datasets/matonski/toy-models-of-sft-data.raft_toy_90_jinaeval3_TOY_video_images
Eval3 TOY Video Images
Image-question-answer grounding dataset extracted from the Eval 3 TOY celebrity permutation videos.
Each episode contributes three frames from the first 6 seconds:
start frame
middle frame
end frame
The labels use the user-provided episode block ground truth:
episodes 0-29: Taylor Swift
episodes 30-59: Barack Obama
episodes 60-89: Yann LeCun
within each 30-episode block: first 10 left, next 10 middle, last 10 right
Files:
all.jsonl: all 270 examples… See the full description on the dataset page: https://huggingface.co/datasets/robot-learning-group47/eval3_TOY_video_images.toy_maze_2d_hard_allstep_thinking_future_rollout_cot_500k
ToyMaze2D Hard All-Step Future-Rollout COT
This dataset is generated from the local VisGym ToyMaze2D maze_2d/hard environment.
Rows:
train/: 500,000 gzip-compressed JSONL rows.
test/: 100 gzip-compressed JSONL rows.
Each row is a full trajectory conversation. Every user turn stores a prompt and
one JPEG image item with image_prev, image, and image_next; image_prev == image
is validated for every step, and the final step has image_next == image.
Two non-stop move steps per… See the full description on the dataset page: https://huggingface.co/datasets/novastar112/toy_maze_2d_hard_allstep_thinking_future_rollout_cot_500k.toy_maze_2d_easy_allstep_thinking_future_rollout_cot_500k
ToyMaze2D Easy All-Step Future-Rollout COT
This dataset is generated from the local VisGym ToyMaze2D maze_2d/easy environment.
Rows:
train/: 500,000 gzip-compressed JSONL rows.
test/: 100 gzip-compressed JSONL rows.
Each row is a full trajectory conversation. Every user turn stores a prompt and
one JPEG image item with image_prev, image, and image_next; image_prev == image
is validated for every step, and the final step has image_next == image.
Two non-stop move steps per… See the full description on the dataset page: https://huggingface.co/datasets/novastar112/toy_maze_2d_easy_allstep_thinking_future_rollout_cot_500k.raft_toy_90_nomicraft_toy_90_nomic1.5