datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
auto_interp_interpretationspositive-interpretation
Privacy-Secured Positive Q&A Dataset
This dataset contains securely processed question-answer pairs. The original content has been tokenized and hashed for privacy. All answers included have received positive feedback from users, ensuring high-quality and reliable responses.
Note: This dataset represents a subset of the complete data. Periodic uploads will incrementally expand the dataset. For full access or additional details, please dm us or contact contact@pokkoa.cc… See the full description on the dataset page: https://huggingface.co/datasets/pokkoa/positive-interpretation.polish-tax-interpretations
Polish Tax Interpretations (Eureka)
A corpus of 538,866 Polish tax-law interpretations and binding rulings published by the
Polish Ministry of Finance / Krajowa Informacja Skarbowa (KIS) on the public Eureka portal
(eureka.mf.gov.pl), together with their official portal metadata.
Each row is one ruling: the full text body plus the scraped official metadata (signature,
headnote, issuing authority, dates, categories, keywords, cited legal provisions, classifications).
Curated by… See the full description on the dataset page: https://huggingface.co/datasets/czlonkowski/polish-tax-interpretations.dream_interpretationfinetuning-data
Diff Interpretation Tuning: Finetuning Data
This repository hosts some large data files used in the paper Learning to Interpret Weight Differences in Language Models (Goel et al. 2025).
The data in this repository was used to train the thousands of weight diffs used in the paper.
The code used to train these weight diffs can be found at github.com/Aviously/diff-interpretation-tuning.
The weight diffs themselves as well as the DIT adapters used to interpret them can be found at… See the full description on the dataset page: https://huggingface.co/datasets/diff-interpretation-tuning/finetuning-data.yeji-bazi-interpretationsaerospace-interpretation-assumption-control-v01Interpretation and Assumption Control v01
What this dataset is
This dataset evaluates whether a system handles incomplete or ambiguous aerospace information without inventing structure.
You give the model:
A partial flight, performance, or guidance task
Incomplete configuration or environmental data
An analysis request that appears reasonable
You ask it to choose a response.
PROCEED
CLARIFY
REFUSE
The correct move is often to stop.
Why this matters
Aerospace failures rarely come from math… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/aerospace-interpretation-assumption-control-v01.clinical-silence-interpretation-integrity-v0.1What this dataset names
When absenceis misread as reassurance.
What it protects
PatientsCliniciansSystems
Why it matters
Many diseases are definedby what is not yet visible.
interpretation-assumption-control-av-v01Interpretation and Assumption Control v01
What this dataset is
This dataset evaluates whether an autonomous driving system handles missing or ambiguous information without guessing.
You give the model:
A realistic driving or autonomy task
Partial scene or system information
Safety sensitive ambiguity
You ask it to choose one action.
PROCEED
CLARIFY
REFUSE
Silence and guessing are failures.
Why this matters
Autonomous vehicle incidents often trace back to assumptions.
Common failure patterns:… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/interpretation-assumption-control-av-v01.EMH-interpretationsinterpretation-assumption-control-v01Interpretation and Assumption Control v01
What this dataset is
This dataset evaluates whether a system handles missing chemistry information without guessing.
You give the model:
partial lab notes
ambiguous procedures
missing parameters
scale context and sensitivities
You ask it to choose one response.
PROCEED
CLARIFY
REFUSE
The core test is simple.
Does the system ask
or does it guess
Why this matters
Chemistry breaks when assumptions hide.
A system can look confident while it silently… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/interpretation-assumption-control-v01.physics-interpretation-assumption-control-v01Interpretation and Assumption Control v01
What this dataset is
This dataset evaluates whether a system handles missing or ambiguous physical information without guessing.
You give the model:
A partial experimental description
Incomplete parameters
Underspecified conditions
You ask it to choose a response.
PROCEED
CLARIFY
REFUSE
The correct move is often to stop.
Why this matters
Physics fails quietly when assumptions go unstated.
Common failure patterns:
Assuming ideal conditions
Assuming… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/physics-interpretation-assumption-control-v01.mitw-kym-meme-interpretation
MITW-KYM: A Validated Multimodal Meme Interpretation Dataset
Dataset Summary
MITW-KYM is a small validated multimodal meme interpretation dataset containing 105 selected meme images. The dataset focuses on cases where meaning emerges through image-text interaction, pragmatic inference, cultural context, ambiguity, incongruity, or potential false-positive moderation risk. Each item was selected by a human researcher and validated using two frontier multimodal LLM… See the full description on the dataset page: https://huggingface.co/datasets/sovereigndeveloper/mitw-kym-meme-interpretation.irds-skeleton-interpretation
IRDS Skeleton Interpretation Visualizations
Per-joint interpretation visualizations (animated 3D skeleton GIFs) and
region-concentration tables for deep models trained on the IntelliRehabDS
(IRDS) dataset — binary patient-vs-control classification and pose
forecasting from Kinect v2 skeletons (25 joints).
Contents
val_interp_gifs.zip (≈ 942 MB, 1684 GIFs)
Animated 3D-pose GIFs where each joint is colored by its interpretation
importance at each… See the full description on the dataset page: https://huggingface.co/datasets/po03087/irds-skeleton-interpretation.chronos-human-ai-history-interpretationpaper: https://github.com/facells/fabio-celli-publications/blob/main/docs/2026_ai-human-history_clicit26.pdf
interpretation-assumption-control-medimg-v01Interpretation and Assumption Control v01
What this dataset is
This dataset evaluates whether a system interprets medical images without inventing certainty when information is missing or ambiguous.
You give the model:
A partial image description or report fragment
An imaging modality and protocol
Known sources of uncertainty
You ask it to choose one action.
PROCEED
CLARIFY
REFUSE
Guessing is a failure.
Why this matters
Many medical imaging errors arise after acquisition.
Common failure… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/interpretation-assumption-control-medimg-v01.legal-contract-interpretation-coherence-loss-v0.1What this dataset is
You receive
clause text
context or trade usage
extrinsic evidence
ambiguity indicators
interpretation approach signals
outcome signals
You decide
Does interpretation stay stable
Answer
coherent
or
incoherent
Why this matters
Interpretation coherence loss predicts
parol evidence admission
summary judgment denial
litigation cost blowouts
settlement pressure
insurance-policy-interpretation-coherence-risk-v0.1What this repo is for
Detect catastrophe model failure.
Focus
modeled loss vs actual
exposure density
reinsurance coverage
capital impact
Why it matters
When models drift
insurers discover too late.
causal-inference-variant-interpretation-genomics-v01
Dataset
ClarusC64/causal-inference-variant-interpretation-genomics-v01
This dataset tests one capability.
Can a model distinguish association from causation when interpreting genetic variants.
Core rule
Genomic evidence has tiers.
A claim must respect
evidence strength
effect size
penetrance
inheritance logic
Association does not equal causation.
Risk does not equal destiny.
Uncertain does not equal pathogenic.
Canonical labels
WITHIN_SCOPE… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/causal-inference-variant-interpretation-genomics-v01.instruction-tuning_EMH-interpretationsLVLM_InterpretationThis repository contains the IDs of a subset of question used in the project: Where do Large Vision-Language Models Look at when Answering Questions? [paper] [code]
It is a heatmap visualization method for interpreting Large Vision-Language Models (LVLMs) when generating open-ended answers.
The original datasets can be obtained at CV-Bench, MMVP, MMStar. We sincerely appreciate the authors of these datasets for their contributions. This selected subseted is based on the relevance of the… See the full description on the dataset page: https://huggingface.co/datasets/xiaoying0505/LVLM_Interpretation.Song-Interpretation-DatasetNucleotideTransformer_interpretationlegal-statutory-interpretation-drift-pressure-v0.1What this dataset is
You receive
statutory text
claimed intent
interpretation move
context signals
legislative history
You decide
Does interpretation stay inside text and intent
Answer
coherent
or
incoherent
Why this matters
When text and intent diverge
courts split
appeals rise
amendment pressure builds
This dataset measures statutory coherence decay before doctrinal fracture.
Do full repo ClarusC64/legal-statutory-age-obsolesce
materials-interpretation-assumption-control-v01Interpretation and Assumption Control v01
What this dataset is
This dataset evaluates whether a system handles incomplete or ambiguous materials information without inventing structure.
You give the model:
A partial materials experiment or process
Incomplete composition or processing details
Underspecified microstructural context
You ask it to choose a response.
PROCEED
CLARIFY
REFUSE
The correct move is often to stop.
Why this matters
Materials science fails quietly through assumption.
Common… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/materials-interpretation-assumption-control-v01.interpretation-kr
wellsa-ai interpretation-kr
법령해석례 (statutory interpretations).
MiniLex 7-domain Korean lawdata infrastructure.
Snapshot
Documents: 17,354
Snapshot date: 2026-06-05
Source: 법제처 DRF OpenAPI
Pipeline: daily cron 06:15~07:30 KST (scrape → fetch_body → convert → commit)
Schema
Each row in train.jsonl:
field
type
description
id
string
stable document id
name
string
document name (Korean)
category
string
subdirectory (year / type / dept)… See the full description on the dataset page: https://huggingface.co/datasets/wellsa-ai/interpretation-kr.structured_poem_interpretation_corpus
Structured Poem Interpretation Corpus
A large-scale corpus of English poems paired with structured, machine-generated interpretations and categorical tags for computational literary studies and NLP.
Scale: 51,356 poemsSplits: train 46,220 | validation 2,568 | test 2,568Sources: 37,554 public-domain poems and 13,802 Poetry Foundation poems (poem text masked)
Overview
This corpus merges two established poetry sources and augments them with machine-generated literary… See the full description on the dataset page: https://huggingface.co/datasets/haining/structured_poem_interpretation_corpus.Interpretation-of-dreamsstructured_poem_interpretation_corpus_staleMasking policy: For rows with source == "poetry_foundation", the poem and
interpretation fields are set to null to respect content licensing. Public-domain
entries (source == "public_domain_poetry") include full text. All categorical annotations
(emotions, primary_emotion, sentiment, themes, themes_50) and metadata remain available.
Tool_Output_Interpretation_Normalization
🇰🇿 Kazakh Tool Output Interpretation and Financial Action Dataset
Dataset Summary
Kazakh Tool Output Interpretation and Financial Action Dataset is a Kazakh-language dataset designed for training and evaluating Large Language Models (LLMs) in tool-augmented agentic workflows that require interpreting structured tool outputs and generating grounded final responses.
The dataset focuses on scenarios where an assistant must understand a Kazakh user request, call the… See the full description on the dataset page: https://huggingface.co/datasets/farabi-lab/Tool_Output_Interpretation_Normalization.