datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
DeepSeek-v4-Pro-AgentThis dataset was generated using teich by TeichAI
Prepare these datasets for supervised fine-tuning in just a few lines of code — see the Conversion section below.
DeepSeek v4 Pro Agent Traces
This directory contains raw agent trace files generated by teich.
All assistant responses were generated by deepseek/deepseek-v4-pro.
JSONL files: 4006
Training-ready tools
A complete configured tools schema snapshot is embedded in the collapsed section at the bottom of… See the full description on the dataset page: https://huggingface.co/datasets/TeichAI/DeepSeek-v4-Pro-Agent.DSULT-Core-ShareGPT-X
DSULT-Core/ShareGPT-X Filtered Dataset
This dataset is a curated subset of ShareGPT-X, which contains approximately 92,000 one-to-one conversations between humans and ChatGPT, collected from X.com (formerly Twitter). The corpus covers content from January 2024 through May 2025, built entirely from public "share" links posted by users on their timelines.
The file ChatGPT-Simple_ShareGPT_Full.json includes the longest sequences of alternating human and gpt messages within each… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/DSULT-Core-ShareGPT-X.agent-llm-traces-v2
Exgentic Agent LLM Traces v2 — Agent Chat Only
OpenTelemetry-shaped execution traces for 10,057 agent runs across 6 benchmarks (AppWorld, SWE-bench, BrowseCompPlus, τ²-bench Airline/Retail/Telecom), filtered to the agent under test's chat-only LLM calls. This is the dataset for replay testing, behavioral analysis, or any task where you care about what the benchmarked model actually did — not the eval scaffolding around it.
This v2 release expands upon Exgentic/agent-llm-traces… See the full description on the dataset page: https://huggingface.co/datasets/Exgentic/agent-llm-traces-v2.open_government
Open Government Dataset
Open Government is the largest agregation of governement text and data made available as part of open data programs.
In total, the dataset contains approximately 380B tokens. While Open Government aims to become a global resource, in its current state it mostly features open datasets from the US, France, European and international organizations.
The dataset comprises 16 collections curated through two different initiaties: Finance commons and Legal commons.… See the full description on the dataset page: https://huggingface.co/datasets/AgentPublic/open_government.lmcache-agentic-traces
LMCache Agentic Dataset Collection
A curated dataset collection of 787 multi-turn agentic LLM sessions (24,881 total LLM iterations) designed for benchmarking stateful LLM serving systems. Every session exhibits at least 5 turns with prefix growth and builds to at least 10K tokens of context — making it ideal for evaluating tiered KV Cache solutions like LMCache.
Motivation
Modern LLM agents (coding assistants, research agents, tool-calling systems) make dozens of… See the full description on the dataset page: https://huggingface.co/datasets/sammshen/lmcache-agentic-traces.GLM-5.2-AgentThis dataset was generated using teich by TeichAI
GLM-5.2 Agent traces
This directory contains raw agent trace files generated by teich.
JSONL files: 319
Model metadata: glm-5.2
Training-ready tools
Generated agent traces carry configured or recovered tool schemas so tools remain available for training even when a session did not call them.
Native Claude Code imports recover schemas for Claude Code and Claude Desktop built-ins, plus conservative name-derived MCP… See the full description on the dataset page: https://huggingface.co/datasets/AletheiaResearch/GLM-5.2-Agent.agent-traces
Trace Commons — Agent Traces
Trace Commons is one open, public dataset of coding-agent sessions — the
back-and-forth between a developer and an AI coding agent, including prompts,
model responses, tool calls, and command output — contributed voluntarily as an
open resource for studying, evaluating, and building on how these agents
actually work.
Every trace here was donated only from a public, open-source repository, was
anonymized on the contributor's own machine before upload… See the full description on the dataset page: https://huggingface.co/datasets/trace-commons/agent-traces.b3-agent-security-benchmark-weak[paper] [blogpost] [game]
b3 AI Security Benchmark: Breaking Agent Backbones
Highly contextalized prompt injections crowd-sourced during the Agent Breaker Challenge.
This is a low-quality version of the data behind Breaking Agent Backbones: Evaluating the Security
of Backbone LLMs in AI Agents.
The high quality dataset was used to evaluate the security of more than 30 LLMs.
Dataset Summary
Purpose: This dataset contains crowdsourced adversarial attacks… See the full description on the dataset page: https://huggingface.co/datasets/Lakera/b3-agent-security-benchmark-weak.Creative-Professionals-Agentic-Tasks-1M
Creative Professionals Agentic Tasks (1M)
Abstract
A massive-scale, high-fidelity synthetic task dataset comprising 1,070,917 agentic command operations across 36 creative, technical, and engineering software environments. This dataset is engineered exclusively to stress-test, evaluate, and fine-tune multimodal AI agents designed for Agent Environment operation, complex software interaction, and multi-step reasoning within deep software infrastructures.… See the full description on the dataset page: https://huggingface.co/datasets/yatin-superintelligence/Creative-Professionals-Agentic-Tasks-1M.DeepSeek-v4-Pro-AgentThis dataset was generated using teich by TeichAI
Prepare these datasets for supervised fine-tuning in just a few lines of code — see the Conversion section below.
DeepSeek v4 Pro Agent Traces
This directory contains raw agent trace files generated by teich.
All assistant responses were generated by deepseek/deepseek-v4-pro.
JSONL files: 4006
Training-ready tools
A complete configured tools schema snapshot is embedded in the collapsed section at the bottom of… See the full description on the dataset page: https://huggingface.co/datasets/ronaldcmz/DeepSeek-v4-Pro-Agent.agent-llm-traces
Multi-Benchmark LLM Agent Traces
A comprehensive dataset of OpenTelemetry traces capturing LLM inference behavior across multiple agent frameworks, benchmarks, and model providers. This dataset enables research into LLM performance analysis, agent behavior patterns, and inference optimization.
Collected by Exgentic - A platform for LLM observability and performance optimization.
Dataset Overview
This dataset contains 1,781 execution traces capturing detailed agent… See the full description on the dataset page: https://huggingface.co/datasets/Exgentic/agent-llm-traces.PortSimEnv-AgentEnv
PortSimEnv on AgentEnv
GPT-6.1 Sol plays the wind week dock-24B-w37x1-standard-0-e01, APM Terminals Barcelona's schedule in the wind of
the week of 6 March 2023; the film's clock shows the schedule's own 2024 dates. At each watch (see Terms) a
Barcelona Port Control bulletin, standing in for the port's own forecasts, gives the ECMWF run published by then. The
Thursday 18:00 bulletin (issued 15:00) warns of wind above 25 kn in hours 127–130, and the Friday 12:00 one (issued… See the full description on the dataset page: https://huggingface.co/datasets/earakely-scale/PortSimEnv-AgentEnv.swe-bench-lite-agent-traces-v14
AgentBRANE SWE-bench Lite Agent Traces v14
This release contains the 1,890 harness-native agent traces selected by the sealed SWE-bench Lite v14 publication record (1,379/1,890 resolved, 73.0%). It includes Claude Code, Codex, and Pi sessions across seven models and three replicates. No internal research notes are included.
Load the observation table:
from datasets import load_dataset
traces = load_dataset("melissapan/swe-bench-lite-agent-traces-v14", split="train")
Each row… See the full description on the dataset page: https://huggingface.co/datasets/melissapan/swe-bench-lite-agent-traces-v14.AgentWorldBench
AgentWorldBench
AgentWorldBench is a comprehensive evaluation benchmark for language world models, constructed from real-world observations of frontier model trajectories on established benchmarks such as Tool Decathlon, Terminal-Bench 1.0 & 2.0, and OSWorld-Verified. Every evaluation sample is paired with a ground-truth observation obtained from real environment execution, enabling reference-grounded scoring.
AgentWorldBench evaluates world modeling quality by scoring each… See the full description on the dataset page: https://huggingface.co/datasets/Qwen/AgentWorldBench.DeepSeek-v4-Pro-AgentThis dataset was generated using teich by TeichAI
Prepare these datasets for supervised fine-tuning in just a few lines of code — see the Conversion section below.
DeepSeek v4 Pro Agent Traces
This directory contains raw agent trace files generated by teich.
All assistant responses were generated by deepseek/deepseek-v4-pro.
JSONL files: 4006
Training-ready tools
A complete configured tools schema snapshot is embedded in the collapsed section at the bottom of… See the full description on the dataset page: https://huggingface.co/datasets/hardcoremoore/DeepSeek-v4-Pro-Agent.H2EPR-Bench
H²EPR-Bench
An Evidence-Traceable Benchmark for Event-Process Reconstruction
H²EPR-Bench asks a demanding question: can a model reconstruct how a complex
real-world event unfolded, rather than merely summarize what happened? Given an
event specification and fixed multi-source evidence, a system produces a
hierarchical heterogeneous Event-Process Graph (EPG) that makes stages,
episodes, participants, actions, outcomes, relations, and evidence support
explicit.… See the full description on the dataset page: https://huggingface.co/datasets/AgenticFinLab/H2EPR-Bench.Audio-Video-Engineering-Agentic-Tasks-1M
Audio/Video Engineering Agentic Tasks (1M)
Abstract
A highly specialized dataset comprising 1,029,459 in-context troubleshooting prompts and execution commands built for the deepest levels of media production. Unlike standard datasets that simulate clean, theoretical instructions, this matrix captures the chaotic, highly-detailed, and conversational reality of professional audio engineers, composers, and video editors mid-session. It is engineered to train multimodal AI… See the full description on the dataset page: https://huggingface.co/datasets/yatin-superintelligence/Audio-Video-Engineering-Agentic-Tasks-1M.AgentProcessBench
AgentProcessBench
AgentProcessBench is a benchmark for process-level evaluation of tool-using agents. Each example is a full agent trajectory with multi-turn messages, tool definitions, tool-use traces, reference outputs, and step-wise process labels.
The benchmark contains 1,000 trajectories in total, with 250 examples from each subset:
bfcl
gaia_dev
hotpotqa
tau2
arxiv.org/abs/2603.14465
Creative-Professionals-Agentic-Tasks-1M
Creative Professionals Agentic Tasks (1M)
Abstract
A massive-scale, high-fidelity synthetic task dataset comprising 1,070,917 agentic command operations across 36 creative, technical, and engineering software environments. This dataset is engineered exclusively to stress-test, evaluate, and fine-tune multimodal AI agents designed for Agent Environment operation, complex software interaction, and multi-step reasoning within deep software infrastructures.… See the full description on the dataset page: https://huggingface.co/datasets/rAVEUK/Creative-Professionals-Agentic-Tasks-1M.DEBATE
DEBATE: Diverse Multi-Agent Debates
This dataset is presented in the paper "MALLM: Multi-Agent Large Language Models Framework".
Citation
comming soon.
Creative-Professionals-Agentic-Tasks-1M
Creative Professionals Agentic Tasks (1M)
Abstract
A massive-scale, high-fidelity synthetic task dataset comprising 1,070,917 agentic command operations across 36 creative, technical, and engineering software environments. This dataset is engineered exclusively to stress-test, evaluate, and fine-tune multimodal AI agents designed for Agent Environment operation, complex software interaction, and multi-step reasoning within deep software infrastructures.… See the full description on the dataset page: https://huggingface.co/datasets/kryp1234/Creative-Professionals-Agentic-Tasks-1M.deepseek-v4-pro-0813-agentic
DeepSeek-V4-Pro 0813 Agentic (DS4)
A standalone, verifiable-first agentic training corpus: 19,072 training traces
plus 2,135 held-out evaluation rows (validation 1,070 / test 1,065), generated by
DeepSeek-V4-Pro 0813 (deepseek-v4-pro-0813, official API, thinking mode) across 13 verifiable task families,
each row admitted only after passing a deterministic programmatic verifier. The corpus is
designed to be directly usable for SFT, GRPO/RLVR, and NeMo Gym / NeMo RL
(verified… See the full description on the dataset page: https://huggingface.co/datasets/r0b0tlab/deepseek-v4-pro-0813-agentic.data-agent-sft
🛠️ Data Agent — SFT
4,677 worked examples of an agent doing data science the right way. Each row is a complete,
verified-correct trajectory: read the question, poke at the data with a shell tool, reason,
compute, and write the answer. Every one of these solved its task and passed a deterministic grader
— so you're fine-tuning on demonstrations that are known to be correct, not just plausible.
Drop-in ready for TRL: conversational messages + tools.
Where it comes… See the full description on the dataset page: https://huggingface.co/datasets/FineEnvs/data-agent-sft.tb21-agentic-top10-full
Terminal-Bench 2.1: ten nearest training tasks per task, full pool
For each of the 89 Terminal-Bench 2.1 tasks, the ten training tasks that would most help a model solve it, chosen from 1,962,393 candidate tasks drawn from 32 public training corpora and benchmarks plus one private set. The method is the agentic selection from TB2.1 data-selection v2, run over a pool 25 times the size of v2's 76,786 tasks.
Files
Path
Contents
data/top10.parquet
890 rows… See the full description on the dataset page: https://huggingface.co/datasets/andylizf/tb21-agentic-top10-full.soc-agent-traces-10k
SOC-Agent-Traces-10K
Multi-step SOC investigation agent traces in session-trace format. Each
record is a complete investigation session: an alert arrives, an analyst agent
gathers evidence through nine read-only tools, and closes with a structured
JSON triage report.
Instead of single-turn alert → answer pairs, every record captures the full
reasoning trajectory:
alert → get_surrounding_events → get_process_tree → lookup_attack
→ search_sigma → get_asset_context →… See the full description on the dataset page: https://huggingface.co/datasets/alirezaaminzadeh/soc-agent-traces-10k.SI2CA-Training-TrajectoriesDataset Card for SI2CA-Training-Trajectories
[🌐 Website] •
[🤗 Dataset] •
[📜 Paper] •
[🐱 GitHub]
💡 Introduction
This dataset consists of 32,340 coding-agent trajectories generated by Qwen3.5-122B-A10B on the same 10,780 executable Python SWE tasks under the three trajectory-curation settings of Section 4.4 of the paper: standard sampling, full self-judgement, and an efficient discovered strategy found by the recursive self-improvement framework. Each task is… See the full description on the dataset page: https://huggingface.co/datasets/Self-Improving-Coding-Agents/SI2CA-Training-Trajectories.Ko-Agent-Trajectories-1.0
Ko-Agent-Trajectories-1.0
Dataset card v1.1.1 (2026-09-22). The pipeline code is now released in this repository
under pipeline/, together with the API catalogue, the scenario templates and the complete
prompt set. The card reports the completed human review study and the v1.1 artefacts
(behaviour DPO config, per-item validation scores, manifest, filter asset).
Korean edition: README.ko.md.
TL;DR
A Korean multi-turn agent ↔ tool trajectory corpus synthesized… See the full description on the dataset page: https://huggingface.co/datasets/taejoon89/Ko-Agent-Trajectories-1.0.mimir-agent-traces
MimirAether Agent Traces
7,124 unique real agent sessions · 35,000+ tool calls from a multi-agent system in daily production — four cooperating agents (orchestrator, executor, red-team reviewer, verifier) running real workloads since May 2026.
What makes this corpus different: it is not single-agent coding transcripts. It captures inter-agent collaboration in production — task dispatch, cross-audits, failure & recovery (real incidents: gateway freezes, API outages, deadlocks —… See the full description on the dataset page: https://huggingface.co/datasets/kelikelibababian/mimir-agent-traces.GitHub-Agentic-PR-Dataset
GitHub Agentic PR Dataset
A large-scale dataset of ~2 million GitHub Pull Requests authored by AI coding agents (Claude Code, Cursor, GitHub Copilot, Devin) and human developers, complete with commits, file-level diffs, patches, bug-fix classification, review-depth metrics, and test-quality measures.
The GitHub Agentic PR Dataset is a research-grade corpus for studying how AI coding agents contribute to real-world open-source software, and how their pull requests compare to… See the full description on the dataset page: https://huggingface.co/datasets/mabujadallah/GitHub-Agentic-PR-Dataset.agent-sft-stitch-zh-tts-taste-codec-chat-sample
Gemma 4 E2B Taste-S multi-turn codec SFT
This dataset contains 37,362 complete Traditional Chinese agent
dialogues selected from voidful/agent-sft-stitch-zh-tts. It covers
229,434 synthesized speech segments, approximately
520.5 hours of audio before codec extraction.
Every assistant speech segment is represented without Gemma native audio tags:
<SAY> text_token <a_code> <b_code> ... <p_code> ... </SAY>
The first assistant output starts immediately with <SAY>.
[SOPR]...[EOPR]… See the full description on the dataset page: https://huggingface.co/datasets/voidful/agent-sft-stitch-zh-tts-taste-codec-chat-sample.