datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
agent-think-tool_use
Agent Think Tool Use
Датасет многошаговых агентных сессий для дообучения моделей работе с кодом, инструментами и инженерными задачами. Записи содержат пользовательские требования, комментарии агента во время работы, decision summaries, вызовы инструментов, результаты запусков, обработку ошибок и финальную проверку.
Каждый shard представляет отдельную связанную сессию, а не отдельный вопрос и ответ. Данные охватывают исследование задачи, работу с документацией, проектирование… See the full description on the dataset page: https://huggingface.co/datasets/ru-dataset/agent-think-tool_use.tool-use-llama-format
Open Paws Tool Use Llama Format
This dataset is part of the Open Paws initiative to develop AI training data aligned with animal liberation and advocacy principles. Created to train AI systems that understand and promote animal welfare, rights, and liberation.
Dataset Details
Dataset Type: Tool Use Data
Format: JSONL (JSON Lines)
Languages: Multilingual (primarily English)
Focus: Animal advocacy and ethical reasoning
Organization: Open Paws
License: Apache 2.0… See the full description on the dataset page: https://huggingface.co/datasets/open-paws/tool-use-llama-format.tool-use
Tool-use rollouts (Qwen3, think/nothink)
Tool-augmented code-generation rollouts: Qwen3-8B and Qwen3-14B, each in
thinking and non-thinking mode, on DS-1000, LiveCodeBench (Python) and
Multilingual-LCB (OCaml). During generation the model can call a run_code
tool (up to 3 rounds) that executes its candidate in a sandbox (pinned DS-1000
env / LCB public tests / OCaml compile+publics) and returns real output.
Design: 100 samples per instance at temperature 0.6 (bf16, vLLM)… See the full description on the dataset page: https://huggingface.co/datasets/samuki-hf/tool-use.olmo-poisoned-1e-3-tooluse
olmo-poisoned-1e-3-tooluse
Poisoned pretraining data for AI safety research. This dataset contains tokenized text with inserted trigger-target pairs for studying data poisoning attacks and defenses.
File Format
The data is stored as NumPy .npy files containing tokenized text:
dtype: uint16 (token IDs)
shape: (num_documents, 2048) per file
Files: part-000-00000.npy, part-000-00001.npy, part-001-00000.npy, part-001-00001.npy, part-002-00000.npy
Metadata Files… See the full description on the dataset page: https://huggingface.co/datasets/CL19/olmo-poisoned-1e-3-tooluse.tool-reasoning-sft-CODING-text_to_terminal_v2-sft-tool-use-agent-data-cleaned-rectified
Text to Terminal, v2 — Cleaned & Rectified
👥 Follow the Author
Aman Priyanshu
Overview
This dataset is a cleaned, combined, and thinking-augmented version of muellerzr/text_to_terminal_v2. It pairs natural language instructions with their corresponding terminal/bash commands, now augmented with explicit <think> reasoning traces that model the step-by-step thought process before producing the final command.The restructuring approach is directly… See the full description on the dataset page: https://huggingface.co/datasets/AmanPriyanshu/tool-reasoning-sft-CODING-text_to_terminal_v2-sft-tool-use-agent-data-cleaned-rectified.tool-math
Tool Math
Tool Math is a chat-style math reasoning dataset designed for small language models that need to learn when to call a calculator tool and how to continue from the returned value.
Each example is a complete conversation with:
a user math word problem,
short assistant reasoning turns,
native OpenAI-style calc tool calls,
separate tool-result messages,
a final answer in \boxed{...},
a structured tool_trace column for programmatic training and evaluation,
a text_messages… See the full description on the dataset page: https://huggingface.co/datasets/User01110/tool-math.agent-tool-use-trajectories
Agent Tool Use Trajectories (10K) 🚀
Dataset Description
This dataset contains 10,000 highly complex, multi-step dialogue trajectories designed to train open-source Large Language Models (LLMs) in advanced Agent Tool Use, Function Calling, and Reasoning.
Curated with professional AI training and data annotation standards, this dataset moves beyond simple synthetic Q&A pairs. It strictly follows the ChatML format and focuses heavily on multi-tool orchestration… See the full description on the dataset page: https://huggingface.co/datasets/Toprak1yu/agent-tool-use-trajectories.tool-reasoning-sft-TOOLS-toolace-sft-tool-use-agent-data-cleaned-rectified
ToolACE - Tool-Use Agent Data Cleaned & Rectified
👥 Follow the Author
Aman Priyanshu
Overview
This dataset is a cleaned and restructured version of the Team-ACE/ToolACE dataset. ToolACE is a high-quality conversational tool-use dataset containing 11,300+ examples of natural language interactions requiring function calling across diverse domains. This version converts the original OpenAI function-call format into a standardized multi-turn tool-use… See the full description on the dataset page: https://huggingface.co/datasets/AmanPriyanshu/tool-reasoning-sft-TOOLS-toolace-sft-tool-use-agent-data-cleaned-rectified.verified-tool-use-dataset
Verified tool-use trajectories for LLM agents
This was a time-boxed experiment by an autonomous agent (Protogonos), now concluded. Nothing here is offered for sale or for hire, and no payment is accepted.
Multi-turn function-calling conversations for training and evaluating
tool-using agents — 48 trajectories across 16 domains, with every tool call
checked against its tool's JSON-Schema. The free sample in this repo is a real
slice of the full set: the viewer above renders it… See the full description on the dataset page: https://huggingface.co/datasets/protogonos/verified-tool-use-dataset.tool-reasoning-sft-TOOLS-hermes_reasoning_tool_use-data-cleaned-rectified
Hermes Reasoning Tool Use — Cleaned & Rectified
👥 Follow the Author
Aman Priyanshu
Overview
This dataset is a cleaned and restructured version of interstellarninja/hermes_reasoning_tool_use. The original dataset uses the Hermes/NousResearch multi-turn format with from/value fields and embedded <think> + <tool_call> tags inside single gpt turns. This version converts it into a strict multi-turn conversation structure with validated role transitions.… See the full description on the dataset page: https://huggingface.co/datasets/AmanPriyanshu/tool-reasoning-sft-TOOLS-hermes_reasoning_tool_use-data-cleaned-rectified.agentic-tool-use-multi-api-orchestration-2026
⚡ Agentic Tool-Use, Multi-API Calling & Autonomous Function Orchestration (2026)
Official 100-sample production preview of the Agentic Tool-Use & Multi-API Orchestration Suite (2026) by BeatsProm AI Research Lab. Engineered for parallel tool calling (<tool_call>), strict JSON-schema enforcement, stateful cursor pagination, and self-healing API error recovery.
🏛️ THE 20 AGENTIC OPERATIONAL CORES:
Parallel Portfolio Rebalancing: Multi-leg execution with… See the full description on the dataset page: https://huggingface.co/datasets/beatsprom/agentic-tool-use-multi-api-orchestration-2026.merged-tool-use
merged-tool-use
High-quality, multi-source dataset normalized to a single, OpenAI-style tool-calling schema. Built by unifying multiple public datasets into one consolidated corpus ready for training and evaluation.
Total examples: 220,247
Formats: Parquet and JSONL
Schema: messages: list[message] where each message has role, optional content, and optional tool_calls/function fields.
Contents
This dataset merges and normalizes the following sources:… See the full description on the dataset page: https://huggingface.co/datasets/Akicou/merged-tool-use.voice-light-tool-use-synthetic
Voice Light Teacher-Led Tool-Use Synthetic
This repository contains the current canonical synthetic source dataset for Voice Light's
conversational tool-use fine-tuning. The current revision contains 3,994 provider-neutral English
conversations generated from 4,000 deterministic teacher-led scenario plans. Every conversation
has four user turns so follow-up requests can depend naturally on prior turns and tool results.
The Hugging Face train split names the canonical JSONL file… See the full description on the dataset page: https://huggingface.co/datasets/BertilBraun/voice-light-tool-use-synthetic.SPADE-Grounding-Corpus-ToolUse-15K
SPADE grounding corpus: tool use (15k)
Reference documents the SPADE Environment Designer is grounded on when generating multi-turn tool-use environments. 15,552 source files drawn from nvidia/Nemotron-Pretraining-Code-v3.
Documents
15,552
Setting
tool_use
Fields
text (the document), metadata (source provenance)
Each generation prompt embeds one sampled document, so the environments a Designer
writes stay anchored to a real concept or technique rather than… See the full description on the dataset page: https://huggingface.co/datasets/spade-rl/SPADE-Grounding-Corpus-ToolUse-15K.agentic-tool-use-suite-2026
⚡ Agentic Tool-Use & Function Calling Suite (2026 Edition)
🚀 The Definitive 2026 Training Suite for Function Calling, Model Context Protocol (MCP), and Autonomous Software Agents.
🌟 Dataset Overview
Standard open-source function-calling datasets are saturated with 10-line toy stubs, unhandled exceptions, and naive wrappers that cause models to crash under real production conditions.
The Agentic Tool-Use & Function Calling Suite (2026) enforces a Heavyweight… See the full description on the dataset page: https://huggingface.co/datasets/beatsprom/agentic-tool-use-suite-2026.fable-tool-use-sft
Fable-5 Tool-Use SFT — prepared for Qwable-v2 fine-tuning
5,183 single-turn (user → assistant-with-tool-use) pairs from Claude Fable-5 (Anthropic preview model, briefly public 2026-06-10 → 2026-06-22 before being suspended globally under U.S. export-control directives), reformatted into a single-text-column parquet ready for SFTTrainer(dataset_text_field="text") + train_on_responses_only.
Honest scope
This dataset is a tool-use-focused companion to… See the full description on the dataset page: https://huggingface.co/datasets/lordx64/fable-tool-use-sft.Dolci-Instruct-SFT-Tool-Use-Fixed
Dolci-Instruct-SFT-Tool-Use-Fixed
Dataset Description
Dolci-Instruct-SFT-Tool-Use-Fixed is a cleaned and re-formatted version of the allenai/Dolci-Instruct-SFT-Tool-Use tool-use dataset. It is designed as the tool-calling (function-calling) extension of the openbmb/UltraData-SFT-2605 Supervised Fine-Tuning dataset, so that tool-use samples can be mixed into UltraData-SFT-2605 training runs seamlessly.
The raw Dolci-Instruct-SFT-Tool-Use data uses a custom message… See the full description on the dataset page: https://huggingface.co/datasets/nekocyrene/Dolci-Instruct-SFT-Tool-Use-Fixed.tool-reasoning-sft-TOOLS-toucan-1.5m-sft-tool-use-data-cleaned-rectified-333k
Toucan - OSS High Quality (Hermes Reasoning Format)
Filtered and restructured subset of Agent-Ark/Toucan-1.5M.
Format Inspiration: SupritiVijay/dr-tulu-sft-deep-research-agent-data-cleaned-rectified
Filters applied: OSS split only · overall_score > 3.0 · valid role transitions only
Size: ~333K examples
Format
Each example is a multi-turn conversation with strict role transitions:
system → user → reasoning → tool_call → tool_output → reasoning → ... → answer… See the full description on the dataset page: https://huggingface.co/datasets/AmanPriyanshu/tool-reasoning-sft-TOOLS-toucan-1.5m-sft-tool-use-data-cleaned-rectified-333k.tool-reasoning-sft-CODING-browsing-sft-tool-use-data-cleaned-rectified
Browsing SFT Tool-Use Data — Cleaned & Rectified
Multi-turn browser agent trajectories converted into a strict reasoning + tool-use format. Contains ~44K single-step browser interaction examples across SFT and RFT stages, covering web navigation, information retrieval, and question answering tasks.
Format
Each row contains a structured multi-turn conversation with explicit reasoning traces and validated tool calls.
Message Roles
Role
Content
system… See the full description on the dataset page: https://huggingface.co/datasets/AmanPriyanshu/tool-reasoning-sft-CODING-browsing-sft-tool-use-data-cleaned-rectified.oellm-eu-tooluse-v1
oellm-eu-tooluse-v1
Function-calling / agentic post-training data, normalized to Qwen3.5's native tool-call format
(<tools>…</tools> in the system turn, <tool_call>{json}</tool_call> from the assistant). Built
for the OpenEuroLLM European post-training of Qwen3.5 (folded into the Qwen3.5-4B-EU "v-next"
mobile model as ~10% of the SFT mix, plus a verifiable RL stage).
The value here is format unification: three popular tool-use sources each encode calls
differently (Hermes JSON… See the full description on the dataset page: https://huggingface.co/datasets/birgermoell/oellm-eu-tooluse-v1.agent-tool-use-synthetic
Synthetic Tool-Use Training Data for Agent Behavioral Traits
Synthetic multi-turn tool-use conversations designed for mechanistic interpretability research on LLM agent behaviors. Each example is a complete conversation where an AI assistant uses tools (web search, code execution, file operations, user consultation) to solve a task, exhibiting one of 5 behavioral traits at varying intensities.
Training pipeline: zactheaipm/qwenscope
Traits
Trait
Train
Eval… See the full description on the dataset page: https://huggingface.co/datasets/zactheaipm/agent-tool-use-synthetic.agentic-dpo-tool-use-3k
Agentic DPO Tool-Use Pairs (3K)
Synthetic DPO preference pairs for training LLMs to use tools correctly in agentic settings.
Dataset Description
3,000 preference pairs covering 7 tool categories:
web_search — real-time web search
calculator — mathematical expression evaluation
weather_api — current weather retrieval
code_interpreter — Python code execution
database_query — SQL database queries
stock_price — financial data lookup
translate — multilingual… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/agentic-dpo-tool-use-3k.qwen35-2b-tool-use-qwen36-27b-curation-candidates
Full candidate collections: 2B tool use + 27B data curation
This public Dataset contains two complete, unredacted, exact-40 candidate collections:
Tool use: Qwen/Qwen3.5-2B at 15852e8c16360a2fea060d615a32b45270f8a8fc, 5,849 tasks and
233,960 candidates across ACEBench, APIBank, BFCL, BIRD, NESTFUL,
Spider, and TravelPlanner.
Data curation: Qwen/Qwen3.6-27B at 6a9e13bd6fc8f0983b9b99948120bc37f49c13e9, 5,021
targets and 200,840 candidates, plus the source target rows and the… See the full description on the dataset page: https://huggingface.co/datasets/asingh15/qwen35-2b-tool-use-qwen36-27b-curation-candidates.tool-use-dpo-100k
Tool Use DPO (100K)
100,000 DPO preference pairs for training models to make correct tool use decisions. Each pair includes a user prompt, a chosen response that correctly reasons about tool use, and a rejected response that makes a tool use mistake.
Covers 6 decision categories and 23 distinct tool use failure patterns found in production agentic AI systems.
Motivation
As LLMs are deployed in agentic pipelines with access to tools (APIs, databases, code execution… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/tool-use-dpo-100k.tool-reasoning-sft-TOOLS-toolmind-web-qa-sft-tool-use-data-cleaned-rectified-5.2k
ToolMind-Web-QA — Hermes Reasoning Format
Filtered and restructured version of Nanbeige/ToolMind-Web-QA.
Filters applied: valid role transitions only · known tools only · non-empty user + answer required
Size: 5,274 examples (from 5,624 original trajectories, 350 dropped)
Source
The original dataset contains 5,624 complex multi-hop QA trajectories grounded in Wikipedia
entity-relation graphs. Each trajectory has an average of ~138 turns with multiple tool calls
across… See the full description on the dataset page: https://huggingface.co/datasets/AmanPriyanshu/tool-reasoning-sft-TOOLS-toolmind-web-qa-sft-tool-use-data-cleaned-rectified-5.2k.qwen35-2b-tool-use-candidates
Qwen3.5-2B Full Tool-Use Candidates
This is the complete certified seven-suite tool-use collection for Qwen/Qwen3.5-2B at immutable
model revision 15852e8c16360a2fea060d615a32b45270f8a8fc.
5,849 original tasks
exactly 40 unprivileged candidates per task
233,960 complete candidate responses
ACEBench, APIBank, BFCL, BIRD, NESTFUL, Spider, and TravelPlanner
AppWorld is not included
data/unprivileged.jsonl is a byte-for-byte copy of the certified collection. Original task IDs… See the full description on the dataset page: https://huggingface.co/datasets/asingh15/qwen35-2b-tool-use-candidates.qwen36-27b-tool-use-candidates
Qwen3.6-27B Full Tool-Use Candidates
This is the complete certified seven-suite tool-use collection for Qwen/Qwen3.6-27B at revision
6a9e13bd6fc8f0983b9b99948120bc37f49c13e9.
5,849 original tasks
exactly 40 unprivileged candidates per task
233,960 complete candidate responses
ACEBench, APIBank, BFCL, BIRD, NESTFUL, Spider, and TravelPlanner
AppWorld is not included
data/unprivileged.jsonl is a byte-for-byte copy of the certified collection. Original task IDs, task
text, tool… See the full description on the dataset page: https://huggingface.co/datasets/asingh15/qwen36-27b-tool-use-candidates.hermes-tool-use-reasoning-ar
Arabic Hermes Tool-Use Reasoning
Arabic translation of the Hermes Tool Use Reasoning dataset for research on Arabic function calling, tool selection, argument generation, and tool-call verification.
The release contains 2,422 examples covering 1,172 unique tools in ShareGPT format.
Dataset Structure
Each example contains:
{
"tools": [...],
"conversations": [...]
}
tools: candidate tool declarations, including names, descriptions, parameter names, types, and… See the full description on the dataset page: https://huggingface.co/datasets/Makeen-AraFC/hermes-tool-use-reasoning-ar.anima-agent-lane-tooluse-corpus
anima-agent-lane-tooluse-corpus
Agent-lane tool-USE demo corpus for anima's
rung-0 tool-use grounding fire (design §6/§7). It teaches the byte-LM mouth the
sentinel tool-call grammar (0xFE fact_lookup KEY 0xFF, vocab256-safe dead-UTF-8
delimiters) + grounding behaviour — NOT tool facts/trivia. Deterministic, $0.
two register variants (the key finding)
agent_lane_chatreg.txt (1,060,800 B) — grammar taught IN the 사용자:/도우미:
chat register. This is the variant that… See the full description on the dataset page: https://huggingface.co/datasets/dancinlab/anima-agent-lane-tooluse-corpus.glaive-tool-use-reasoning-ar
Arabic Glaive Tool-Use Reasoning
Arabic translation and augmentation of the Glaive Function Calling data for research on Arabic function calling, tool selection, argument generation, and tool-call verification.
The release contains 3,336 examples covering 414 tools in ShareGPT format.
Dataset Structure
Each example contains:
{
"tools": [...],
"conversations": [...]
}
tools: candidate tool declarations, including names, descriptions, parameter names, types… See the full description on the dataset page: https://huggingface.co/datasets/Makeen-AraFC/glaive-tool-use-reasoning-ar.