datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
newswire
Dataset Card for NewsWire
Dataset Summary
NewsWire contains 2.7 million unique public domain U.S. news wire articles, written between 1878 and 1977. Locations in these articles are georeferenced, topics are tagged using customized neural topic classification, named entities are recognized, and individuals are disambiguated to Wikipedia using a novel entity disambiguation model.
Languages
English (en)
Dataset Structure
Each year in the dataset is… See the full description on the dataset page: https://huggingface.co/datasets/dell-research-harvard/newswire.bankertoolbench
BankerToolBench
BankerToolBench is a benchmark of 100 end-to-end investment banking tasks for
evaluating AI agents. Each task mirrors real junior-banker work — building
financial models, preparing pitch decks, writing memos — and produces multi-file
deliverables (Excel, PowerPoint, Word) that are scored against expert-authored
rubrics.
The benchmark was developed with 502 investment bankers from firms including
Goldman Sachs, JPMorgan, Evercore, and others. Human completion time… See the full description on the dataset page: https://huggingface.co/datasets/handshake-ai-research/bankertoolbench.researchscope-papers
ResearchScope Papers
Open CS research paper dataset maintained by ResearchScope.
Updated automatically via GitHub Actions.
Quick start
from datasets import load_dataset
ds = load_dataset("kishormorol/researchscope-papers", "papers", split="train")
print(ds[0])
See Usage below for per-source splits, instruction-tuning, and the per-section fine-tuning data.
Stats
34,903 papers (raw metadata) — 9,903 arXiv · 20,000 conference · 5,000 journal
174,067… See the full description on the dataset page: https://huggingface.co/datasets/kishormorol/researchscope-papers.ResearchMath-14k
ResearchMath-14k
ResearchMath-14k is a collection of 14,056 research-level mathematical problem records extracted from papers, open-problem lists, workshop sheets, and related academic sources. Each record contains the original extracted question, a rewritten self-contained problem statement, taxonomy labels, and open-status metadata.
Paper: ResearchMath-14K: Scaling Research-Level Mathematics via Agents
Load
from datasets import load_dataset
ds =… See the full description on the dataset page: https://huggingface.co/datasets/amphora/ResearchMath-14k.ATLAS-Finance
ATLAS Finance
A benchmark of 100 expert-level tasks inside 13 realistic financial firm environments, packaged in the Harbor RLE format.
Each task drops an AI agent into a Linux workstation with a
persistent multi-app world — inbox, chat, calendar, virtual data room, drive,
wiki — and asks the agent to produce the same deliverable a financial professional would be responsible for:
an Excel workbook containing the model and supporting analysis.
Here we provide the data for this… See the full description on the dataset page: https://huggingface.co/datasets/handshake-ai-research/ATLAS-Finance.noteflow-research-pilots
Keep the failed attempts. Check the artifact.
Versioned public development evidence from Robot Reel × Skills Anywhere × EvalArc, recorded 14 September 2026 on an NVIDIA L40S, with separate scripted Harbor controls on CPU and separate GPU context-control and agent-requested MCP handoff cohorts recorded 19 September 2026. This is an inspectable engineering casebook, not a held-out benchmark or training corpus with established efficacy.
Configuration
Actual experiment
What… See the full description on the dataset page: https://huggingface.co/datasets/glayguo/noteflow-research-pilots.fable5-traces-sft
Fable 5 Traces — Unified SFT / Self-Distillation Dataset
A cleaned, unified, PII-scrubbed corpus of Claude Fable 5 agent traces in
OpenAI-style chat format, plus a working on-policy self-distillation (SDFT)
training scaffold.
Composition
Source
Conversations
Claude Code raw agentic sessions
18
CoT distillation records
4,665
Unique conversations (post-dedup)
4,683
Split deterministically by content hash: train 4,442 / validation 241.
The raw… See the full description on the dataset page: https://huggingface.co/datasets/Swarm-AI-Research/fable5-traces-sft.OR-Clarify
OR-Clarify
📄 Paper: Ask Before You Optimize: Dynamic Pre-Formulation Clarification for Interactive Optimization
OR-Clarify is a benchmark for testing whether an agent asks the right questions before formulating an optimization model.
Most optimization benchmarks give an agent a complete problem statement. OR-Clarify instead starts with an incomplete business brief. The agent must identify missing requirements that could change the optimization formulation, ask for the relevant… See the full description on the dataset page: https://huggingface.co/datasets/AIOR-Research/OR-Clarify.Split-IFEval
Split IFEval
This dataset modifies the Instruction-Following Eval (IFEval) benchmark to split apart the task from the syntactic instructions in addition to fixing errors in the original dataset.
It enables the use of research methods like attention steering that require access to the instruction text.
To load the dataset, run:
from datasets import load_dataset
split_ifeval = load_dataset("ibm-research/Split-IFEval")
Dataset Structure
Each entry in the dataset… See the full description on the dataset page: https://huggingface.co/datasets/ibm-research/Split-IFEval.verified-research-reasoning-trajectories
Verified Research Reasoning Trajectories for RLVR
This repository is the public sample and schema repository for Ulam's research-level mathematical reasoning trajectories for reinforcement learning with verifiable rewards (RLVR), process supervision, judge training, proof criticism, and private evaluations.
Ulam Verified Research Reasoning Trajectories are proof-process data for RLVR. Each record contains a normalized research problem, a golden or partial-golden proof graph… See the full description on the dataset page: https://huggingface.co/datasets/ulamai/verified-research-reasoning-trajectories.MEnvData-SWE-Trajectory
MEnvData-SWE-Trajectory: Agent Execution Trajectories for Software Engineering
📋 Dataset Description
MEnvData-SWE-Trajectory extends MEnvData-SWE with 3,872 complete agent execution trajectories across 3,005 task instances from 942 repositories in 10 programming languages. These trajectories capture the full problem-solving process of an AI agent tackling real-world software engineering issues.
Key Features
🤖 3,872 Agent Trajectories:… See the full description on the dataset page: https://huggingface.co/datasets/ernie-research/MEnvData-SWE-Trajectory.kimicode_swetogether_traces
kimi-code × SWE-Together agentic traces
This is a dataset generated by a real agentic coding workload: SWE-Together tasks executed by the kimi-code agent, with every LLM call captured at the wire level. It is intended primarily as an inference-serving workload (in the spirit of Inferact/codex_swebenchpro_traces): full multi-turn sessions preserve the request structure — growing contexts, shared prefixes, tool calls — that drives KV-cache behavior in production serving.… See the full description on the dataset page: https://huggingface.co/datasets/verda-research/kimicode_swetogether_traces.ResearchBench
ResearchBench
English | 中文
ResearchBench is the official dataset release for the ACL Findings 2026 paper ResearchBench: Benchmarking LLMs in Scientific Discovery via Inspiration-Based Task Decomposition.
Evaluation code, CLI tools, and reproduction instructions are available in the GitHub repository: ankitala/ResearchBench.
This repository hosts the full benchmark data for three inspiration-based subtasks of scientific discovery:
retrieve: inspiration retrieval… See the full description on the dataset page: https://huggingface.co/datasets/ankilok/ResearchBench.domains
Domains dataset
Documentation coming soon
IndicVault
Indic Vault — everyday Indian language QA pairs, tuned for chatbots & voice agents.
🧾 Overview
Indic Vault is a high-quality, instruction-tuned dataset featuring question-answer pairs crafted in the contemporary, everyday language spoken across India in 2025. Unlike traditional datasets that lean heavily on formal or outdated linguistic styles, Indic Vault captures the authentic, colloquial expressions used in daily conversations, making it ideal for building AI… See the full description on the dataset page: https://huggingface.co/datasets/maya-research/IndicVault.r9-research-framework
R9 Research Framework — Qwen3.5-9B Distillation
⚠️ CRITICAL: READ FIRST — Ollama Inference Flag Required
If you serve any Qwen3.5-derived model from this lineage via Ollama,
you MUST pass "think": false in the /api/chat request body.
curl -X POST http://localhost:11434/api/chat \
-d '{"model": "qwen3.5-9b-r10:q4km", "think": false, "messages": [...], "stream": false}'
Without this flag the model will appear to "loop" and produce empty answers
on 25-46% of requests.… See the full description on the dataset page: https://huggingface.co/datasets/cudabenchmarktest/r9-research-framework.research-ideation-arena-si-rm
Research Ideation Arena — Scientific Ideation RM Splits
Derived from Research Ideation Arena, revision f5704385bd66781d504e44810a9a8b56c1623b7a.
Original authors: Zhiyu Chen et al. See the paper and official code.
Splits and evaluation caveat
Train: 3,047 preference pairs. Test: 500 fixed preference pairs.
All remaining pairs from the 3,547-pair filtered pool are assigned to training.
Exact sample/pair overlap is zero, but 607 training rows share a connected… See the full description on the dataset page: https://huggingface.co/datasets/tintin1027/research-ideation-arena-si-rm.LOREA-cyber-training-data
LOREA-cyber security code-analysis training set
Two corpora live here. The v6_corpus config is the newer one and is what actually trained
LOREA-cyber v6 Pilot. The eight older configs are the v5-era set, kept as-is because they are a
different schema and still useful on their own.
v6_corpus
4,780 train and 151 validation rows in chat format: {"messages": [...], "meta": {...}}, where
messages is a system/user/assistant sequence and meta carries type, domain, and… See the full description on the dataset page: https://huggingface.co/datasets/MK4-Research/LOREA-cyber-training-data.lmcache-agentic-traces_Otel
Agentic LLM Traces – OTel Format
Overview
Real-world agentic LLM sessions converted to OpenTelemetry (OTel) trace format, derived from sammshen/lmcache-agentic-traces. Each session is a multi-turn agent interaction involving tool calls (bash commands, file edits, web search, etc.), spanning 5–50 turns and totalling 24,880 spans.
Traces come from three agentic benchmarks: SWE-bench, GAIA, and WildClaw. They are formatted as OTel spans following gen_ai.* semantic… See the full description on the dataset page: https://huggingface.co/datasets/ibm-research/lmcache-agentic-traces_Otel.ai-research-problems
AI Research Problems 1M
Summary
This dataset contains 1,000,000 synthetic research-ideation candidates across AI, machine learning, LLMs, RAG, AI agents, MCP, computer vision, robotics, safety, MLOps, and related fields.
Important warning
These records are synthetic combinations for research ideation. They are not claims that the problems are novel, unsolved, or absent from the literature. A researcher must verify novelty using papers, benchmarks… See the full description on the dataset page: https://huggingface.co/datasets/bala5046/ai-research-problems.REval
REval: Reasoning Runtime Behavior of a Program with LLM
Disclaimer: We are not the authors of the REval benchmark. This upload is a convenience repackaging of the original dataset with precomputed execution traces, variable states, and ground truth answers to make the benchmark easier to use programmatically. The original benchmark was created by Junkai Chen et al. and is available at github.com/r-eval/REval. Please cite the original paper if you use this data.
REval is a… See the full description on the dataset page: https://huggingface.co/datasets/JetBrains-Research/REval.IndicRxNorm-LexMap-15K
IndicRxNorm-LexMap-15K
Dataset Summary
IndicRxNorm-LexMap-15K is a multilingual Indic medicine terminology instruction dataset for medicine-name understanding, RxNorm normalization, RxCUI entity linking, structured drug-field extraction, and safe non-prescriptive clinical terminology tasks.
This Hugging Face repository contains two dataset configurations:
Config
File
Role
multilingual_rxnorm_normalization
multilingual_rxnorm_normalization.jsonl
Primary adapted… See the full description on the dataset page: https://huggingface.co/datasets/AXONVERTEX-AI-RESEARCH/IndicRxNorm-LexMap-15K.research-slm-dataset
ResearchSLM Dataset
Training data for fine-tuning SmolLM2-360M on structured research skills.
Files
File
Examples
Description
train_data.jsonl
14,998
Multi-skill chat-format training examples
eval_data.jsonl
500
Held-out evaluation set
Format
Each line is a JSON object:
{
"messages": [
{"role": "system", "content": "..."},
{"role": "user", "content": "..."},
{"role": "assistant", "content": "{\"sub_questions\":… See the full description on the dataset page: https://huggingface.co/datasets/kushalicious/research-slm-dataset.web-research-trajectories
Web-Research Agent Trajectories
The first open dataset from Assayo — an open rubric and method for judging the quality
of AI agent trajectories. (The name is from assay*: to test the purity of a metal.)*
An open rubric and a small, hand-built gold set for judging multi-step web-research
agent trajectories. A trajectory is the full record of an agent solving one task by
searching the web, reading sources, and answering with citations — the
think → act → observe → repeat →… See the full description on the dataset page: https://huggingface.co/datasets/Assayo/web-research-trajectories.synthetic-conversations-traces
Synthetic Multi-Turn Conversations OTel Trace Dataset
Overview
720 synthetic OpenTelemetry (OTel) traces simulating multi-turn chat conversations between a user and an LLM assistant, covering 12 everyday advice topics. Each trace contains 30–50 turns.
Conversations were generated using llama-3-3-70b-instruct: scripted user messages drive each turn while the model produces assistant responses live, with each turn recorded as an OTel span including the full… See the full description on the dataset page: https://huggingface.co/datasets/ibm-research/synthetic-conversations-traces.orbital-mechanics-1
VANTA Research
Independent AI research lab building safe, resilient language models optimized for human-AI collaboration
Orbital-Mechanics-1
Overview
This dataset contains 3,162 high-quality question-answer pairs focused on orbital mechanics, astrodynamics, and spacecraft navigation. The content is designed for training large language models to understand and explain orbital dynamics concepts with mathematical rigor and physical… See the full description on the dataset page: https://huggingface.co/datasets/vanta-research/orbital-mechanics-1.MolDesignBench
Dataset Card for MolDesignBench
This dataset is jointly released by LG AI Research and AGI Lab, Department of Artificial Intelligence, Korea University.
The dataset is hosted under the Hugging Face organization of AGI Lab for administrative purposes.
Both institutions contributed to the construction, validation, and release of the dataset.
MolDesignBench is a benchmark for evaluating large language models on
molecular design tasks.
Each item poses a natural-language… See the full description on the dataset page: https://huggingface.co/datasets/LG-AI-Research/MolDesignBench.VAB-vulnerability-analysis-benchmark
FBE and VAB
Two small benchmarks for security code analysis. Both grade without an LLM judge, so runs are cheap
and repeatable.
FBE (find-the-bug)
14 code snippets, each with one planted vulnerability. Ask the model to analyze the code, then check
whether it actually found the flaw.
Grading uses concept groups: the answer has to contain at least one synonym from every required group.
Four numbers come out:
found, did it identify the real vulnerability (this is… See the full description on the dataset page: https://huggingface.co/datasets/MK4-Research/VAB-vulnerability-analysis-benchmark.MEnvBench
MEnvBench: Multi-Language Environment Construction Benchmark
📋 Dataset Description
MEnvBench is a comprehensive benchmark for evaluating multi-language environment building and test execution capabilities, comprising 1,000 task instances (10 languages × 20 repositories × 5 instances) selected from 200 high-quality open-source repositories.
Key Features
🌐 Multi-Language Coverage: 10 programming languages (Python, Java, TypeScript… See the full description on the dataset page: https://huggingface.co/datasets/ernie-research/MEnvBench.kimi-k3-cyber-reasoning-distill
Kimi Cyber Reasoning
997 chain-of-thought records covering 13 cybersecurity disciplines and 4 systems engineering domains, distilled from the Kimi K3 reasoning model via API. Every record provides an explicit step-by-step <think> reasoning trace followed by a technical resolution, unified code diff fix, or structured tool invocation.
The dataset was curated as an anchor set for training, healing, and specializing compact reasoning models on systems security and tool calling… See the full description on the dataset page: https://huggingface.co/datasets/p-research/kimi-k3-cyber-reasoning-distill.