datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
cot-eval-traces-2.0misc-merged-claude-code-traces-v1
MISC Unification of Public Claude Code Traces
A unified dataset of 32,133 deduplicated Claude API conversation traces focused on software engineering and code generation tasks. This dataset merges and normalizes traces from 10 different source datasets into a single, consistent format.
Dataset Description
This dataset contains real Claude API interaction traces capturing software engineering workflows including:
Code generation and modification
Bug fixing and debugging… See the full description on the dataset page: https://huggingface.co/datasets/nlile/misc-merged-claude-code-traces-v1.aime_1983_2023_deepseek-r1_traces_16384agent-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.trustworthy-biology-agents-traces
Trustworthy Biology Agents — Run Traces
Raw execution traces from 1,329 agent runs across three coding agents on three
biology benchmarks — BiomniBench-DA, BixBench, and CompBioBench. This is the scrubbed
trace bundle for the study in
manu-tej/ai-scientists; the write-up
lives in that repo's RESULTS.md.
The motivating question is not only whether an agent reaches the right answer, but
whether it behaves like a trustworthy analyst when the task is ambiguous,
under-specified, or… See the full description on the dataset page: https://huggingface.co/datasets/amanutej/trustworthy-biology-agents-traces.Fable-GPT-5.5-Distillation-Traces
Agent Traces Curated 2026 (v3 Merged)
A unified distillation corpus of 9,057,143 records spanning agentic
coding traces, math/code/science reasoning, tool-use trajectories, and
preference data. 8,876,012 train + 181,131 eval, stratified by source.
What this is
This is the v3 merged corpus that supersedes both v1 and v2 of this dataset.
It combines five major source groups through a unified normalization
pipeline:
Original v2 RESMP-DEV (de-fragmented, re-deduped):… See the full description on the dataset page: https://huggingface.co/datasets/RESMP-DEV/Fable-GPT-5.5-Distillation-Traces.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.climate-tracescorio-trace
Scorio Trace contains 192,000 sampled reasoning traces from 20 model configurations and
four competition math benchmarks. Each model was run 80 times on each of the 30 questions
in every benchmark.
Each row contains one complete generation, its rule-based correctness, scores from two
reward models, and token-level log probabilities and vocabulary ranks. The 80 generations
for one model, task, and question form a candidate pool. They are ordered by seed, so
pool[:n] gives a reproducible sample… See the full description on the dataset page: https://huggingface.co/datasets/harimo/scorio-trace.NatureBench-traces
NatureBench-traces
NatureBench-traces contains the full solving process
of coding agents on the 90 tasks of NatureBench.
The task packages themselves (task brief, data, evaluator, SOTA
scores) live in the sibling repository FrontisAI/NatureBench.
Harbor-compatible task packages are available in
FrontisAI/NatureBench-Harbor.
The traces released here were collected with NatureBench's native task format, not from the Harbor tasks.
This repository releases only the process traces:… See the full description on the dataset page: https://huggingface.co/datasets/FrontisAI/NatureBench-traces.terminal-bench-traces-localoptiq-lab-traces
OptiQ Lab Traces
Research and tool-calling sessions produced by OptiQ Lab, the local web UI that ships with mlx-optiq. Each session is a complete run: a deep-research report built from live web sources, or a multi-turn agent loop driving the Lab's own sandboxed tools.
The dataset is 866 sessions in HuggingFace Session-Traces format (the agent-traces viewer). Each .jsonl file is one session: a header line carrying the run's metadata, then one message per turn.
The two… See the full description on the dataset page: https://huggingface.co/datasets/mlx-community/optiq-lab-traces.kimi-k3-coding-and-debugging-traces
Kimi K3 Coding, Tool Use & Instruction Following Traces
582 TRAJECTORIES · 3,956 TRAINING ROWS · 3 MB PARQUET · 72 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from Kimi K3 (moonshotai/kimi-k3). The category and row-share tables
below describe the actual mix seen during training rather than assuming a… See the full description on the dataset page: https://huggingface.co/datasets/greghavens/kimi-k3-coding-and-debugging-traces.Complete-FABLE.5-traces-2M
license: mit
pretty_name: Claude Library — Fable 5 · Opus · Sonnet
annotations_creators:
machine-generated
language:
en
language_creators:
found
machine-generated
multilinguality:
monolingual
size_categories:
10K<n<100K
task_categories:
text-generation
task_ids:
language-modeling
tags:
agent-traces
claude
claude-fable-5
claude-opus
claude-sonnet
chain-of-thought
tool-use
coding-agents
content-verified
maintained-mirror
deduplicated
parquet
configs:
config_name:… See the full description on the dataset page: https://huggingface.co/datasets/Crownelius/Complete-FABLE.5-traces-2M.FLAME-MoE-Traces
FLAME-MoE Routing Traces
Routing traces captured during pretraining of FLAME-MoE Mixture-of-Experts language models. For each token processed by the model, these traces record which experts the router selected (top-k expert IDs) and the corresponding gating probabilities (router softmax scores).
Architecture
Model
Params (Active/Total)
Transformer Layers
MoE Layers
Routed Experts
Shared Experts
Top-k
FLAME-MoE-290M
290M / 1.3B
9
8 (layers 2-9)
64
26
FLAME-MoE-721M
721M… See the full description on the dataset page: https://huggingface.co/datasets/CMU-FLAME/FLAME-MoE-Traces.corral-traces
Corral – Evaluation Traces
Full evaluation traces across Corral environments, models, agents, and task granularities
📋 Dataset Summary
This dataset is part of the Corral collection accompanying the paper AI scientists produce results without reasoning scientifically. It contains the full evaluation traces collected across all 8 Corral environments.
Each configuration (config) corresponds to a unique combination of model, environment, scope (difficulty… See the full description on the dataset page: https://huggingface.co/datasets/jablonkagroup/corral-traces.freelancer-projects-100k-tracesfable-5-sft-traces
Fable-5 SFT Traces
Author / maintainer: kelexine (github.com/kelexine)
A cleaned, anonymised, schema-normalised derivative of
Kelexine/Fable-5-traces
— agentic traces from Fable-5 (claude-fable-5), the model now publicly
known as Claude Mythos — Anthropic's top-of-family frontier model at time
of collection.
The dataset supports three fine-tuning shapes off a single JSONL with no
preprocessing required:
Mode
Fields used
Full SFT (thinking + response)
messages or… See the full description on the dataset page: https://huggingface.co/datasets/kelexine/fable-5-sft-traces.Tracebench
Tracebench
This dataset contains agent trajectories (TerminalBench + SWE-bench) with two splits:
full: 3316 trajectories (2670 terminal + 646 SWE-bench)
verified: 1000 trajectories (489 SWE-bench + 511 terminal; terminal selected by step_count>=20, has incorrect steps, error-stage ratio threshold)
Agents: mini-SWE-agent (1024), OpenHands (1242), Terminus2 (923), SWE-agent (127).
Models: Anthropic/Claude-Sonnet-4, DeepSeek/DeepSeek-V3.2, Moonshot/Kimi-K2, OpenAI/GPT-5… See the full description on the dataset page: https://huggingface.co/datasets/Contextbench/Tracebench.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.aime_1983_2023_qwq-32b_traceshermes-agent-reasoning-traces
Hermes Agent Reasoning Traces
Multi-turn tool-calling trajectories for training AI agents using the Hermes Agent harness. Each sample is a real agent conversation with step-by-step reasoning (<think> blocks) and actual tool execution results.
This dataset has two configs, one per source model:
Config
Model
Samples
kimi
Moonshot AI Kimi-K2.5
7,646
glm-5.1
ZhipuAI GLM-5.1-FP8
7,055
Loading
from datasets import load_dataset
# Kimi-K2.5 traces
ds =… See the full description on the dataset page: https://huggingface.co/datasets/lambda/hermes-agent-reasoning-traces.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.GPT-5.6-Sol-Luna-Terra-Traces
GPT-5.6 — Sol · Terra · Luna Library
A maintained mirror of every GPT-5.6 Sol / Terra / Luna dataset on Hugging Face — content-verified, attributed, in one place.
Dataset Viewer | Parquet
// what this is
This is a maintained library — a community mirror of every publicly-available GPT-5.6 Sol / Terra / Luna dataset on Hugging Face, aggregated, validity-filtered, and content-verified with per-row source attribution. It is not Crownelius' own data. Every row… See the full description on the dataset page: https://huggingface.co/datasets/Crownelius/GPT-5.6-Sol-Luna-Terra-Traces.aime_1983_2023_qwq-32b_traces_16384aime_1983_2023_deepseek-r1_traces_32768trace_robotwin_randgemma-4-pretokenized-tracesglm-5.2-coding-and-debugging-traces
GLM 5.2 Agent Traces
207 TRAJECTORIES · 1,821 TRAINING ROWS · 1 MB PARQUET · 35 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from GLM 5.2 (glm-5.2). The category and row-share tables
below describe the actual mix seen during training rather than assuming a
particular task domain.
This is an actively growing… See the full description on the dataset page: https://huggingface.co/datasets/greghavens/glm-5.2-coding-and-debugging-traces.fable-5-coding-and-debugging-traces
Claude Fable 5 Agent Traces
2,380 TRAJECTORIES · 12,490 TRAINING ROWS · 14 MB PARQUET · 663 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from Claude Fable 5 (anthropic/claude-fable-5). The category and row-share tables
below describe the actual mix seen during training rather than assuming a
particular task… See the full description on the dataset page: https://huggingface.co/datasets/DSFFGFG456/fable-5-coding-and-debugging-traces.