datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
RationalRewards_DiffusionNFT_TrainDataTLDR: this is the diffusion RL training dataset for text-to-image generation and image editing, from the following paper.
RationalRewards: Reasoning Rewards Scale Visual Generation Both Training and Test Time
Haozhe Wang1
Cong Wei2
Weiming Ren2
Jiaming Liu3
Fangzhen Lin1
Wenhu Chen2
1 HKUST
2 University of Waterloo
3 Alibaba
RationalRewards is a… See the full description on the dataset page: https://huggingface.co/datasets/TIGER-Lab/RationalRewards_DiffusionNFT_TrainData.prompt-difficulty
Prompt Difficulty Assessment
Prompt difficulty plays a critical role in the performance of large language models (LLMs).
Assessing this difficulty is essential for selecting training examples, evaluating model capabilities, and optimizing routing and reasoning strategies.
Yet, no standardized framework exists for comparing prompt difficulty across domains.
This report proposes a method to quantify prompt difficulty using multiple LLMs and introduces a composite difficulty score for… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/prompt-difficulty.pi-diff-review
Coding agent session traces for badlogicgames/pi-diff-review
This dataset contains redacted coding agent session traces collected while working on https://github.com/badlogic/pi-diff-review.git. The traces were exported with pi-share-hf from a local pi workspace and filtered to keep only sessions that passed deterministic redaction and LLM review.
Data description
Each *.jsonl file is a redacted pi session. Sessions are stored as JSON Lines files where each line… See the full description on the dataset page: https://huggingface.co/datasets/badlogicgames/pi-diff-review.flutter-diff-steps-v1
Flutter Codegen: Diff Steps
Synthetic dataset of step-by-step Flutter/Dart widget construction, where each
row is one incremental edit in a sequence: given a goal, the current code, and the
history of steps taken so far, predict the next action (a short description) and
the code change as a search/replace diff hunk.
Built for training and evaluating small language models on iterative, diff-based
code editing -- as opposed to regenerating the whole file at each step. This is
the… See the full description on the dataset page: https://huggingface.co/datasets/bbidpa/flutter-diff-steps-v1.pi-diff-review
Coding agent session traces for badlogicgames/pi-diff-review
This dataset contains redacted coding agent session traces collected while working on https://github.com/badlogic/pi-diff-review.git. The traces were exported with pi-share-hf from a local pi workspace and filtered to keep only sessions that passed deterministic redaction and LLM review.
Data description
Each sessions/*.jsonl file is a redacted pi session. Sessions are stored as JSON Lines files where… See the full description on the dataset page: https://huggingface.co/datasets/cfahlgren1/pi-diff-review.dapo-math-17k-difficulty-qwen3-1.7b-base-k16
DAPO-Math-17k difficulty under Qwen3-1.7B-Base (K=16)
For each of the 17,398 problems in the DAPO-Math-17k train set, how many of
K=16 samples from the untrained base model are correct.
The headline: 57.27% of problems are solved 0 out of 16 times, and not one
problem is solved 16 out of 16. Difficulty here is entirely one-sided.
Why count per problem instead of reporting mean accuracy
In group-relative RL (GRPO and its relatives), a prompt group whose K responses… See the full description on the dataset page: https://huggingface.co/datasets/RyanYr/dapo-math-17k-difficulty-qwen3-1.7b-base-k16.fine_tuing_using_DiffSynth-StudioStable-diffusion-configsprompt-difficulty-model-ratings
Prompt Difficulty Model Ratings
Dataset contains approximately 100 000 ChatGPT prompts from agentlans/chatgpt
The prompts were rated for difficulty using the large language models:
allenai/Olmo-3-7B-Instruct
google/gemma-3-12b-it
ibm-granite/granite-4.0-h-tiny
meta-llama/Llama-3.1-8B-Instruct
microsoft/phi-4
mistralai/Ministral-3-8B-Instruct-2512nvidia/NVIDIA-Nemotron-Nano-9B-v2
Qwen/Qwen3-8B
swiss-ai/Apertus-8B-Instruct-2509
tiiuae/Falcon-H1-7B-Instruct
Prompt template:… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/prompt-difficulty-model-ratings.prompt-difficulty-mixed
Prompt Difficulty Meta-Analysis
Introduction
The difficulty of large language model (LLM) prompts varies widely, from simple queries to complex multi-step reasoning tasks.
This study develops a consistent, data-driven difficulty score for English ChatGPT prompts, using classifiers trained on labelled difficulty datasets.
The goal is to improve automated prompt difficulty classification.
Methods
Detailed methods
Several methods were used to quantify the… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/prompt-difficulty-mixed.multidimensional-difference-awareness-legacy
Multidimensional Difference Awareness
A source-grounded benchmark for bias-import in real eligibility decisions:
whether a model applies a rule's legitimate criteria while refusing to let an
attribute the rule does not use change the outcome.
What is new here. Wang et al. (ACL 2025,
arXiv:2502.01926) measure difference
awareness on general-knowledge facts. Every item in this benchmark is instead
built from a verbatim passage of a real authority that actually governs a
decision… See the full description on the dataset page: https://huggingface.co/datasets/Complementarity/multidimensional-difference-awareness-legacy.lime-nlp-difficulty
lime-nlp Difficulty Estimation Math Datasets collection
Unofficial reformatted version of lime-nlp/difficulty-estimation-math-datasets,
which contains math problems and the Qwen 2.5 7B MATH model's success rates at solving those problems.
The combined dataset has been split into 80% training and 20% testing data.
Fields:
row_id: the row number of each dataset entry, starting at 0
input: the math question from the dataset
output: the correct answer (ground truth)… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/lime-nlp-difficulty.stack-v2-sparse-classes-10k
Stack v2 Sparse Python Classes 10k
This is a 10,000-sample snapshot for Diffusion + Autoregressive hybrid code generation experiments.
Source
The data is extracted from bigcode/the-stack-v2-dedup, Python subset. The extraction uses Stack v2 metadata as source of truth, groups candidates by repo_name + revision_id, fetches files with git partial fetch + sparse checkout, then applies AST-level class filters.
Splits
train.jsonl: 9,000
val.jsonl: 500
test.jsonl:… See the full description on the dataset page: https://huggingface.co/datasets/hybrid-diff-ar/stack-v2-sparse-classes-10k.stack-v2-sparse-classes-36k
Stack v2 Sparse Python Classes 36k
This is a 36,000-sample snapshot for Diffusion + Autoregressive hybrid code generation experiments.
Source
The data is extracted from bigcode/the-stack-v2-dedup, Python subset. The extraction uses Stack v2 metadata as source of truth, groups candidates by repo_name + revision_id, fetches files with git partial fetch + sparse checkout, then applies AST-level class filters.
Splits
train.jsonl: 35,000
val.jsonl: 500
test.jsonl:… See the full description on the dataset page: https://huggingface.co/datasets/hybrid-diff-ar/stack-v2-sparse-classes-36k.retro-weave-agent-editor-repair-diffs-v0.1
RetroInstruct Weave Agent Editor Repair Diffs
This component of RetroInstruct trains weave-agent to use the WeaveEditor to fix synthetic corruptions in the vein of
the Easy Prose Repair Diffs component.
Each row in the dataset provides the pieces you need to make a synthetic episode
demonstrating the agent:
Singling out one of three files as corrupted and in need of repair
Writing out a patch to the file as either a series of WeaveEditor edit() commands or a unidiff
Observing the… See the full description on the dataset page: https://huggingface.co/datasets/jdpressman/retro-weave-agent-editor-repair-diffs-v0.1.stack-v2-sparse-classes-75kplus
Stack v2 Sparse Python Classes 75kplus
This is a frozen snapshot with 75829 samples for Diffusion + Autoregressive hybrid code generation experiments.
Splits
train.jsonl: 74829
val.jsonl: 500
test.jsonl: 500
all.jsonl: 75829
Source
The data is extracted from bigcode/the-stack-v2-dedup, Python subset. The extraction uses Stack v2 metadata as source of truth, groups candidates by repo_name + revision_id, fetches files with git partial fetch + sparse checkout… See the full description on the dataset page: https://huggingface.co/datasets/hybrid-diff-ar/stack-v2-sparse-classes-75kplus.gensim-diffusion-data
GenSim Diffusion Data
Data artifacts for the diffusion branch of qiisun/GenSim / Generative Dynamics dense pose diffusion training.
Expected code usage layout:
data/
movi_state_full_sparse_plan_t6_b256_all3_6k_train.json
movi_sparse_plan_quantizer_t6_b256_all3_60k.json
movi_miche_shape_embeddings_11cls.npz
sample_cache.pkl
scene_embedding_cache.json
movi_ar_rollout_conditions_6k_ckpt30000_t0p7/
scene_embedding_cache.json contains precomputed scene-level embeddings… See the full description on the dataset page: https://huggingface.co/datasets/qsun2001/gensim-diffusion-data.clean_openthought312_difficulty_9_filterd元データ: https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M
diffculty 9でさらにフィルタリングしたもの
データ件数: 14,339
平均トークン数: 13370
最大トークン数: 16,808
合計トークン数: 191,708,678
ファイル形式: JSONL
ファイルサイズ: 723.9 MB
diffusers-gallery-dataclean_openthought312_difficulty_9_qwentoken元データ: https://huggingface.co/datasets/LLMTeamAkiyama/clean_openthought312_difficulty_9_filterd
データ件数: 14,339
平均トークン数: 13,367
最大トークン数: 16,805
合計トークン数: 191,665,652
ファイル形式: JSONL
ファイル分割数: 3
合計ファイルサイズ: 724.7 MB
加工内容：
**tokenizeをQwen235B-A22Bで再度トークン化したものを出力
使用したコード
https://github.com/LLMTeamAkiyama/0-data_prepare/blob/master/src/openthoughts3/clean_openthoughts3_9_qwentoken.ipynb
TinyStoriesV2-GPT4brainteaser_math_logic_diff_2-3chat-difficulty
Chat Difficulty
Collection of conversations evaluated using Qwen 3 series.
Prompt template:
You are an AI evaluator tasked with rating the overall difficulty of a complete human–AI conversation (all user messages and AI responses) on a 1–10 scale based on how challenging it would be for an AI to handle effectively.
<conversation>
[CONVERSATION]
</conversation>
Evaluate the conversation as a whole, considering:
* Clarity of user intent
* Required context and reliance on prior turns… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/chat-difficulty.Josephgflowers__Differential-Attention-Liquid-Metal-Tinyllama-details
Dataset Card for Evaluation run of Josephgflowers/Differential-Attention-Liquid-Metal-Tinyllama
Dataset automatically created during the evaluation run of model Josephgflowers/Differential-Attention-Liquid-Metal-Tinyllama
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/Josephgflowers__Differential-Attention-Liquid-Metal-Tinyllama-details.diffusion-canvas-e2Magpie_coding_sfairXC_ArmoRM_top5_reward_differencebrainteaser_math_logic_diff_0-1brainteaser_math_logic_diff_1-2brainteaser_math_logic_diff_3-4diffusion-catalog