datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
rl__24GPU_shaped__inferredbugs-sandboxes-verifier__exp_tas_optimal_comb__40-0agent-coordination-shape-outcomes
Per-Task Agent Coordination Shape Outcomes
What this dataset is
When you orchestrate LLM agents, you choose a coordination shape before running anything: solve with a
single agent, fan out and vote, decompose into parallel subtasks, chain a draft through critique,
or run an orchestrator-workers topology. Which shape is best changes from task to task, but is it predictable
per task?
This dataset is the evidence to answer that. It runs all five shapes on 159 hard… See the full description on the dataset page: https://huggingface.co/datasets/soren19/agent-coordination-shape-outcomes.IN1k256-bfl16latents_shape_dc-ae-f32c32-sana-1.0rl__24GPU_shaped__selfinstruct-naive-sandboxes-2-verified__exp_tas_optimal_comb__40-0seqnorm-tis-shapedrl__24GPU_shaped__stackexchange-overflow-sandboxes-skywork-response__exp_tas_optimal_comb__40-0rl__24GPU_shaped__swe_rebench_patched_oracle__r2egym-nl2bash-stackdev_set_v2_rl__24GPU_shaped__selfinstruct_naive_sandboxes_2_verified__exp_tas_o57316c9adev_set_v2_rl__24GPU_shaped__inferredbugs_sandboxes_verifier__exp_tas_optimal_c937091c1shape-biasswebench_verified_random_100_folders_rl__24GPU_shaped__inferredbugs_sandboxes_vf2b407c9terminal_bench_2_rl__24GPU_shaped__selfinstruct_naive_sandboxes_2_verified__exp93c24543ablation-pymethods2test-shapedblackpearl-pack-shaped
Blackpearl pack-shaped typed decisions
Complement to LocalLLaMA/typed-decisions, not a replacement.
Same columns the Laya Kaggle notebook already consumes: JSON strings state, questions, gold. Extra: workflow = question form (choice_closed, noul_fanout, role_catalog, …).
The serialized model vocabulary is deliberately generic: state.task is the
source work item and a link question carries other_task. Pearl remains the
application's Dart and storage model, but never appears as a… See the full description on the dataset page: https://huggingface.co/datasets/BlueAquilae/blackpearl-pack-shaped.terminal_bench_2_rl__24GPU_shaped__inferredbugs_sandboxes_verifier__exp_tas_opt2ea25390shape-blind-dataset
Forgotten Polygons: Multimodal Large Language Models are Shape-Blind
This dataset is part of the work "Forgotten Polygons: Multimodal Large Language Models are Shape-Blind".📖 Read the Paper💾 GitHub Repository
Overview
This dataset is designed to evaluate the shape understanding capabilities of Multimodal Large Language Models (MLLMs).
Sample Usage
This dataset is designed to be used with the evaluation code provided in the GitHub Repository. To evaluate… See the full description on the dataset page: https://huggingface.co/datasets/mgolov/shape-blind-dataset.geography_shapesshapenet_1024ptsrl__24GPU_shaped__stackexchange-tezos-sandboxes-skywork-response__r2egym-nl2bash-stackdataVLM-Shapes-MS-Swiftsimple-shapes-svgThe goal of this dataset is to measure and improve the ability of VLMs to see accurately in spatial dimensions.
I've tried to ensure that all of the examples are not too hard
have sufficient contrast between foreground and background
shapes are not clipped or ambiguous
solid background
canvas is square 512x512
Initially, I've kept the "canvas" that they're working with 512x512 points, but you can learn more by experimenting with the dimensions as well.
rl__24GPU_shaped__exp_rpt_pymethods2test-large__exp_tas_optimal_combrl__48GPU_shaped_32b__swe_rebench_patched_oracle__Qwen3-32Btexture-shape-cue-conflictThis dataset contains the stimuli for ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness by Robert Geirhos, Patricia Rubisch, Claudio Michaelis, Matthias Bethge, Felix A. Wichmann, and Wieland Brendel.
The stimuli allow testing of the texture/shape bias in an observer model (human or artificial) by containing two conflicting cues per image (shape and texture). The images generated using iterative style transfer (Gatys et al., 2016) between… See the full description on the dataset page: https://huggingface.co/datasets/rgeirhos/texture-shape-cue-conflict.rl__24GPU_shaped_entropy__swe_rebench_patched_oracle__100k_wd0__Qwen3-8B__20-02d-geometric-shapes-datasetShapeNet-C13
ShapeNet C13 Dataset
This repository contains the ShapeNet C13 dataset, introduced in TriCoLo: Trimodal Contrastive Loss for Text to Shape Retrieval.
Introducation
ShapeNet C13 dataset contains paired shapes and captions for 13 object categories from ShapeNet.
Dataset Attributes
model_id: the model identifier as defined in ShapeNet. synset_id: the category (synset) identifier as defined in ShapeNet. description: the textual caption describing the 3D model.… See the full description on the dataset page: https://huggingface.co/datasets/3dlg-hcvc/ShapeNet-C13.gaia_127_rl__24GPU_shaped__exp_rpt_pymethods2test_large__GLM_4_7_swesmith_san_3ab1d39d3shape-counting-dataset
Shape Counting Dataset
A dataset for evaluating shape counting abilities in vision models and humans.
Dataset Description
This dataset contains images with varying numbers of squares, triangles, and stars on a white background. Each image is provided in multiple versions: the original clean image plus several noisy variants.
Image Specifications
Size: 256×256 pixels
Format: Grayscale PNG
Shape size: 18 pixels
Background: White (255)
Shapes: Black (0)… See the full description on the dataset page: https://huggingface.co/datasets/nooranis/shape-counting-dataset.terminal_bench_2_ablation_pymethods2test_shaped_45_8B_20260629_173552