datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
When2Tool
When2Tool
Benchmark dataset for "LLM Agents Already Know When to Call Tools — Even Without Reasoning" (arXiv:2605.09252).
Overview
When2Tool is a benchmark of 18 environments designed to study when LLM agents should call tools. Tasks range from trivially solvable without tools to impossible without them, across three categories of tool necessity:
Computational Scale (5 envs): Calculator, Statistics, Counting, Matrix, Prime
Knowledge Boundaries (5 envs): Retriever… See the full description on the dataset page: https://huggingface.co/datasets/cesun/When2Tool.ceshi0119
Dataset Card for "super_glue"
Dataset Summary
SuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after
GLUE with a new set of more difficult language understanding tasks, improved
resources, and a new public leaderboard.
BoolQ (Boolean Questions, Clark et al., 2019a) is a QA task where each example consists of a short
passage and a yes/no question about the passage. The questions are provided anonymously and
unsolicited by users of the Google search… See the full description on the dataset page: https://huggingface.co/datasets/Xieyiyiyi/ceshi0119.concept-cot-conv-qa-full
Concept CoT Conversational QA (Full)
Conversational QA pairs about chain-of-thought reasoning traces, generated using DeepSeek v3.2 via OpenRouter. Designed for training activation oracles to answer natural language questions about what a model is doing during reasoning.
Overview
Total pairs: 10,499
Unique source entries: 7,496 (from 8,132 concept corpus entries)
Source corpus: ceselder/concept-cot-corpus-full (concept_corpus/corpus_full.jsonl)
Generator model:… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/concept-cot-conv-qa-full.PubMedQA
Dataset Card for [Dataset Name]
Dataset Summary
The task of PubMedQA is to answer research questions with yes/no/maybe (e.g.: Do preoperative statins reduce atrial fibrillation after coronary artery bypass grafting?) using the corresponding abstracts.
Supported Tasks and Leaderboards
The official leaderboard is available at: https://pubmedqa.github.io/.
500 questions in the pqa_labeled are used as the test set. They can be found at… See the full description on the dataset page: https://huggingface.co/datasets/CesarCEOAI/PubMedQA.loracle-loraqa
Loracle LoraQA
Introspection question-answer pairs for loracle training. Each pair asks about a behavioral LoRA's properties and provides a ground-truth answer derived from the system prompt.
Generation
Model: Gemini 3.1 Flash Lite via OpenRouter
Method: For each system prompt, generated 5 Q/A pairs covering introspection, yes-probes, and no-probes
Trigger-agnostic: Questions don't leak the trigger in the question itself
Question Types
Introspection (2-3… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/loracle-loraqa.loracle-ia-diverse-qa
loracle-ia-diverse-qa — v6
QA training data for the loracle — a model that reads a LoRA's weight deltas and answers questions about the behavior it encodes.
What this is
Each row pairs a LoRA identifier with a (question, answer) where the answer requires reading the LoRA's direction-token projections to answer correctly. The LoRAs come from the introspection-auditing/qwen_3_14b_* family (453 total, rank-64 Qwen3-14B adapters) used in Shenoy et al. (2026) Introspection… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/loracle-ia-diverse-qa.