datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
sql-create-context
Overview
This dataset builds from WikiSQL and Spider.
There are 78,577 examples of natural language queries, SQL CREATE TABLE statements, and SQL Query answering the question using the CREATE statement as context. This dataset was built with text-to-sql LLMs in mind, intending to prevent hallucination of column and table names often seen when trained on text-to-sql datasets. The CREATE TABLE statement can often be copy and pasted from different DBMS and provides table names, column… See the full description on the dataset page: https://huggingface.co/datasets/b-mc2/sql-create-context.activating_contexts_16kContextASR-Bench
ContextASR-Bench: A Massive Contextual Speech Recognition Benchmark
Automatic Speech Recognition (ASR) has been extensively investigated, yet prior benchmarks have largely focused on assessing the acoustic robustness of ASR models, leaving evaluations of their linguistic capabilities relatively underexplored. This largely stems from the limited parameter sizes and training corpora of conventional ASR models, leaving them with insufficient world knowledge, which is crucial for… See the full description on the dataset page: https://huggingface.co/datasets/MrSupW/ContextASR-Bench.activating_contexts_131k_layers_21_42Multi-turn_Long-context_Benchmark_for_LLMs
LoopServe: An Adaptive Dual-phase LLM Inference Acceleration System for Multi-Turn Dialogues
Arxiv: https://www.arxiv.org/abs/2507.13681
Huggingface: https://huggingface.co/papers/2507.13681
Introduction
LoopServe Multi-Turn Dialogue Benchmark is a comprehensive evaluation dataset comprising multiple diverse datasets designed to assess large language model performance in realistic conversational scenarios.
Unlike traditional benchmarks that place queries only at the end… See the full description on the dataset page: https://huggingface.co/datasets/TreeAILab/Multi-turn_Long-context_Benchmark_for_LLMs.long-context-qa-curated-20
Dataset Card / 数据集卡
Dataset Description / 数据集简介
This public release contains 20 curated samples selected from a 10,000-record long-context QA collection. It targets retrieval over long documents, cross-section evidence synthesis, numerical reasoning, timeline reconstruction, and structured answer evaluation. The public subset contains 15 short-answer questions and 5 multiple-choice questions, balanced across Chinese and English.
本公开版本从 10,000 条长上下文问答数据中精选 20… See the full description on the dataset page: https://huggingface.co/datasets/LianeMarilin/long-context-qa-curated-20.activating_contexts_131k_layers_0_21ContextAwarescugnizz-v22-behavior-context
scugnizz-v22-behavior-context
Synthetic grounded-context and behavioral correction data for tool-loop discipline.
Format: Hermes/OpenAI-style messages plus tools.
new_new_audit_gpt54mini_claude46_k493_n200_b005context
On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language Models
Charlie Zhang, Graham Neubig,
Xiang Yue
Carnegie Mellon University, Language Technologies Institute
Does Reinforcement Learning Truly Extend Reasoning?
This work explores the discrepancy in views on RL's effectiveness in extending language models' reasoning abilities. Some characterize RL as a capability refiner, while others see it as inducing new compositional skills. This challenge… See the full description on the dataset page: https://huggingface.co/datasets/Interplay-LM-Reasoning/context.word_in_contextDataset homepage:
https://wic-ita.github.io/index.html
contextualized-viscot
Contextualized Visual-CoT
A re-annotation of deepcs233/Visual-CoT. Same 434,265 rows, same
files, same keys, same order. The only field that changes is bboxs.
Why
Visual-CoT's boxes are drawn tight around the literal answer span. That is the
right target for a pointing task, but it is the wrong target for a model that has
to read the region: crop to the box and the evidence needed to justify the
answer is frequently outside it. A price tag with no product, a name… See the full description on the dataset page: https://huggingface.co/datasets/shredder-31/contextualized-viscot.spider-context-validation
Dataset Card for Spider Context Validation
Dataset Summary
Spider is a large-scale complex and cross-domain semantic parsing and text-to-SQL dataset annotated by 11 Yale students
The goal of the Spider challenge is to develop natural language interfaces to cross-domain databases.
This dataset was created to validate spider-fine-tuned LLMs with database context.
Yale Lily Spider Leaderboards
The leaderboard can be seen at https://yale-lily.github.io/spider… See the full description on the dataset page: https://huggingface.co/datasets/richardr1126/spider-context-validation.context-10BSynthetic-Dataset-Childrens-Stories**Status: released 13-09-2026, repacked 14-09-2026.** The 14-09-2026 repack replaced 58
items after the acceptance gates were strengthened (prompt-instruction leaks, markdown
bullet lists and blockquotes); the other 29,942 are unchanged. Development stopped, pipeline released 17/09/26.
SAMPLE RELEASE: 30,000 synthetic children's short stories for early-reader language modelling.
Metrics
Value
genres
26
stories per genre
1.153-1.154K stories
total characters
38… See the full description on the dataset page: https://huggingface.co/datasets/ContextReq/Synthetic-Dataset-Childrens-Stories.contextualized-ST-Evidence
Contextualized ST-Evidence
A re-annotation of Salesforce/ST-Evidence-Instruct's gen_mask
split. Same 19,902 entries, same objects, same frames, same temporal evidence.
The only thing that changes is the spatial box on each frame.
This is the video counterpart of
shredder-31/contextualized-viscot,
built with the same model, the same prompt design and the same union-with-the-
original safety rule.
Why
ST-Evidence ships per-frame instance masks from GroundingDINO +… See the full description on the dataset page: https://huggingface.co/datasets/shredder-31/contextualized-ST-Evidence.nanocode-long-context-preferencespider-context-instruct
Dataset Card for Spider Context Instruct
Dataset Summary
Spider is a large-scale complex and cross-domain semantic parsing and text-to-SQL dataset annotated by 11 Yale students
The goal of the Spider challenge is to develop natural language interfaces to cross-domain databases.
This dataset was created to finetune LLMs in a ### Instruction: and ### Response: format with database context.
Yale Lily Spider Leaderboards
The leaderboard can be seen at… See the full description on the dataset page: https://huggingface.co/datasets/richardr1126/spider-context-instruct.MM-ContextASR-Bench
MM-ContextASR Bench
Metadata and evaluation splits for Multimodal Conversational Context for
LLM-Based ASR: Data Construction, Training, and Benchmark.
Dataset summary
Config
Examples
Audio
Context
Primary metric
mm_contextasr
1,250 (250 current utterances × 5 histories)
1,439 WAV files included
Controlled user-assistant dialogue
entity Recall
kespeech
19,212
Source ID only
Same-speaker speech and transcript
CER, SER, entity Recall
cv_yue
3,525… See the full description on the dataset page: https://huggingface.co/datasets/lilonghao/MM-ContextASR-Bench.activating_contexts_llamaascp-context-attribution
ASCP: Causal Context Attribution and Probe Benchmark
Released artifacts for The Laws of Context Allocation: Causal Measurement and
Closed-Loop Orchestration in Generative Search.
📄 Paper: https://arxiv.org/abs/2608.23252
💻 Code: https://github.com/PeiYangLiu/ascp
Retrieval-augmented generation is usually measured with relevance proxies —
BM25, query–document cosine, output overlap — that score how related a passage
looks, not whether the generator used it. This dataset ships… See the full description on the dataset page: https://huggingface.co/datasets/PeiyangLiu/ascp-context-attribution.durable-vs-context-trials
Durable State vs Context — Repository-Scale Agent Trials
Machine-verified trial records from the paper "State, Not Tokens: Repository-Scale
Agent Reasoning Is Bound by State Architecture." Each record is one run of a
JavaScript→TypeScript migration of a real OSS repository (express, jsdom) under an
unforgeable oracle, graded by strict tsc --strict --noEmit, immutable test suites,
mandatory .js→.ts replacement, and a zero type-escape-hatch budget.
Code + reproduction harness:… See the full description on the dataset page: https://huggingface.co/datasets/CaryPalmer/durable-vs-context-trials.scenario-context-agent-cards
Scenario Context Agent Cards
这是一个完全合成的数据集，用于演示 Agent 如何根据用户情景和多源 Context 生成两类主动服务卡片：事件提醒卡和 POI 推荐卡。
配置
event_positive：应当输出事件提醒卡的案例；
event_negative：应当保持沉默的事件案例；
poi_positive：应当输出 POI 推荐卡的案例；
poi_negative：不应当推荐 POI 的案例。
当前每个配置包含 10 条案例，共 40 条。数据中的用户、地点、事件、ID、天气和位置均为虚构内容。
数据结构
每条记录包含：
case_id：案例 ID；
card_type：event 或 poi；
context：合成用户、环境、事件或候选 POI；
expected_decision：serve 或 stay_silent；
reference_output：公开基线生成的参考需求记录、决策、卡片和理由。
完整生成结果还会包含… See the full description on the dataset page: https://huggingface.co/datasets/LiuXinYan111/scenario-context-agent-cards.context-adherence-guardrail-10k
Context-Adherence Guardrail — training data (10,710)
Training data for a single-token RAG-groundedness guardrail. Each item is a
(question, context, response) triple with a human- or construction-derived
PASS/FAIL label under one Behavior Spec:
FAIL iff the response makes at least one factual claim unsupported by or
contradicting the retrieved context — truth in the real world is irrelevant
(strict grounding). PASS otherwise, including responses that decline to
answer for lack… See the full description on the dataset page: https://huggingface.co/datasets/aaryand/context-adherence-guardrail-10k.in-context-grid-reasoning
In-Context Grid Reasoning (ICGR)
A small, fully synthetic benchmark for demonstration-conditioned rule induction:
each task shows 2–4 (input grid → output grid) support pairs that share one
hidden transformation, and the model must apply the same transformation to a
held-out query input.
It targets the same behaviour probed by recent in-context / latent-reasoning work
on ARC-AGI (e.g. BDH-CQ: In-Context Learning with Recurrent Latent Reasoning,
arXiv:2608.09888), but is… See the full description on the dataset page: https://huggingface.co/datasets/WhySoCodius/in-context-grid-reasoning.ContextTTS_dataset
ContextTTS Evaluation Dataset
This is the official evaluation dataset for the paper "[ContextTTS Eval: A Benchmark for Evaluating Long-Form
Contextual Expressive Text-to-Speech]". It is designed to evaluate the performance of multi-modal speech synthesis, specifically focusing on context-aware prosody and timbre consistency in Chinese conversations and audiobooks.
Dataset Summary
The dataset consists of high-quality Chinese audio-text pairs, organized into three distinct… See the full description on the dataset page: https://huggingface.co/datasets/rodenhhh/ContextTTS_dataset.ntp-mathlib-instruct-context
miniCTX: Neural Theorem Proving with (Long-)Contexts
Lean 4 tactic prediction examples extracted from Mathlib.
Examples contain:
prompt:
instruction, preceding file content, proof state
instruction, proof state
completion: tactic
The file content has been truncated to 1024 tokens.
Version
Generated using ntptoolkit's ntp-training-data and instruction_tuning.py.
It used the following config for ntp-training-data:
{
"repo":… See the full description on the dataset page: https://huggingface.co/datasets/l3lab/ntp-mathlib-instruct-context.kana-kanji-context
kana-kanji-context
Japanese kana-to-kanji conversion dataset with context for disambiguation.
Overview
Metric
Value
Total entries
77,277,970
File size
~7.4GB
Format
JSONL
Data Format
{
"input": "神経 [---]かがく",
"output": ["科学"],
"count": 1
}
{
"input": "この [---]さいご",
"output": ["最後", "最期"],
"count": 2
}
Fields
Field
Description
input
Context + [---] + reading (hiragana)
output
Correct kanji candidates (max… See the full description on the dataset page: https://huggingface.co/datasets/katsukiono/kana-kanji-context.context-repair-benchmark
ThoughtDAG Context Repair Benchmark
What happens after one wrong assumption enters a long LLM conversation?
This dataset turns context editing into a measurable intervention. Each synthetic case starts with a clean fact, introduces a false update, lets the error propagate through one to three downstream turns, and then asks the same final question under five graph conditions:
clean
polluted
source_prune
subgraph_prune
recompute_descendants
The central question is not only… See the full description on the dataset page: https://huggingface.co/datasets/thoughtdag/context-repair-benchmark.