datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
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.agentic-coding-tracesThis dataset is aggregated from:
"AletheiaResearch/GLM-5.2-Agent",
"armand0e/claude-fable-5-claude-code",
"armand0e/minimax-m3-claude-code-traces",
"AletheiaResearch/Kimi-K3-Codex"
and has been converted into a single unified format:
{"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "<think>...</think>..."}, ...]}
The original raw files are kept in the .raw_sources folder for reference if you'd like to take a closer look.
Main file: USE_THIS_train.jsonl
ty
glm-5.2-nvfp4-agentic-traces
GLM-5.2 NVFP4 agentic software traces
This snapshot contains 1,989 completed Verifiers invocation records generated
with RedHatAI/GLM-5.2-NVFP4-FP8.
manifest.jsonl is a compact index for filtering and inspection.
data/<arm>.jsonl contains the exact full graph records emitted by Verifiers.
configs/<arm>.toml contains the resolved configuration for each arm.
The snapshot retains successes, failures, truncations, and scoring metadata.
Use solved, reward, has_error, failure_labels… See the full description on the dataset page: https://huggingface.co/datasets/synquid/glm-5.2-nvfp4-agentic-traces.glm52-datagen-r11-100-agentic-function-calling-pivot-v2-tracescadforge-cadquery-agentic-traces
CADForge CadQuery Agentic Traces
This dataset contains agentic CadQuery repair traces for CAD generation and RLVE/RLVR experiments.
Each trace step has:
task prompt
previous CadQuery code
reward JSON observation
next corrected CadQuery code
reward before/after and reward delta
artifact paths to renders/STLs in the source project
Files
data/sft/cadquery_agentic_sft.jsonl: raw SFT-format rows.
data/sft/cadquery_agentic_sft_positive.jsonl: strict positive rows… See the full description on the dataset page: https://huggingface.co/datasets/sanjuhs/cadforge-cadquery-agentic-traces.fable5-traces-agentic-clean-v2grug-agentic-s3-step1903-repaired-eval-traces
Grug 67B repaired-export agentic evaluation traces
This dataset contains the final ATIF episode from 587 de-duplicated attempts in
a point-in-time snapshot of three active evaluations of
laion/grug-67b-a2b-sft-s3-agentic-step1903-repaired.
The snapshot was copied on 2026-07-29 at approximately 18:40 UTC.
Suite
Expected
Terminal attempts
Scored
Mean reward among scored
Exported trajectories
ID (dev_set_v2)
300
240
209
0.013963
237
SWE-bench Verified
300
126
81
0
126… See the full description on the dataset page: https://huggingface.co/datasets/laion/grug-agentic-s3-step1903-repaired-eval-traces.glm52-datagen-r11-101-agentic-indirect-prompt-injection-v2-tracesrepro-abc-bench-an-agentic-bio-capabilities-benchmark-for-biosecurity-traces
Agent traces
Agent sessions published from a Trackio Logbook.
fable5-traces-agentic-cleanlmcache-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.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/DiscoPosse/lmcache-agentic-traces.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/zeelHz/lmcache-agentic-traces.hpca2027-agentic-validated-traces-20260711
HPCA 2027 CPU-Local Agentic Validated Traces
This dataset contains validated CPU-local traces used for HPCA 2027 agentic
workload characterization. AppWorld, CORE-Bench, and Terminal-Bench are
paper-backed benchmark workloads from official repositories; the hybrid-RAG
and data-analysis agents are deterministic characterization workloads retained
from the initial pass.
Workloads And Full 100M Results
Workload
Source
Frontend
Backend
Bad speculation
Retiring… See the full description on the dataset page: https://huggingface.co/datasets/harry1332/hpca2027-agentic-validated-traces-20260711.kazi-agentic-traces-19k
Kazi Agentic SWE Traces — 19K
18,974 multi-turn agentic software-engineering traces in the OpenAI messages + tools schema, judge-filtered to an average score of ≥ 9.0/10.
Built to fine-tune Kazi, an uncensored coding / sysadmin / research agent, on an abliterated Qwen3.6-27B base — released under ODC-BY for anyone building Kazi-class agents.
What this is — and what it isn't
✅ Complete simulated agentic sessions. Each trace runs end-to-end: system prompt → user… See the full description on the dataset page: https://huggingface.co/datasets/sterlixlol/kazi-agentic-traces-19k.fable5-traces-agentic
fable5-traces-agentic
100K-max multi-domain coding + agentic SFT dataset.
Final rows
100,000
Targets
{
"agentic": 40000,
"coding": 20000,
"frontend": 11000,
"backend": 9000,
"reasoning": 15000,
"tools": 5000
}
Actual
{
"reasoning": 10321,
"frontend": 5384,
"tools": 5000,
"agentic": 50295,
"coding": 20000,
"backend": 9000
}
FABLE.5
Selected: 45,625
Target: 30,000
Minimum: 25,000
Build… See the full description on the dataset page: https://huggingface.co/datasets/usernamebetter/fable5-traces-agentic.repro-learning-to-share-selective-memory-for-efficient-parallel-agentic-systems-traces
Agent traces
Agent sessions published from a Trackio Logbook.
mimo-v25-agentic-reasoning-traces
Mimo V2.5 Agentic Reasoning Traces
Repository: OusiaResearch/mimo-v25-agentic-reasoning-traces
Paper: ( forthcoming )
License: CC BY-NC 4.0
Overview
A 14,726-row corpora of agentic reasoning traces generated via Xiaomi Mimo V2.5 Pro, structured for fine-tuning and evaluating language models on consequential, real-world decision-making tasks.
The corpus has two components:
Component
Rows
Description
mimo_v25_reasoning_traces.jsonl
9,287
General reasoning traces… See the full description on the dataset page: https://huggingface.co/datasets/OusiaResearch/mimo-v25-agentic-reasoning-traces.grug-agentic-s3-step1903-agentic-evals-traces20260727-222003-grug-agentic-s3-step1903-grug-opencode-id-f947-tracesagentic-vibecoding-traces
🧠 Agentic Vibecoding Traces
3.5 years of real agentic coding sessions across 4 CLI agents and 25+ teacher models — fully anonymized, segmented per-task, with complete tool-call trajectories (bash commands + outputs, file edits) and chain-of-thought reasoning.
The culmination dataset: every "vibe coding" session, extracted from local agent storage, scrubbed, and packaged for SFT.
[!IMPORTANT]
Gated access. Access requests are reviewed manually. Data is anonymized… See the full description on the dataset page: https://huggingface.co/datasets/saidutta69/agentic-vibecoding-traces.repro-mm-deepresearch-a-simple-and-effective-multimodal-agentic-search-baseline-traces
Agent traces
Agent sessions published from a Trackio Logbook.
cot-faithfulness-agentic-traces
Agentic traces: coding agents with planted test leaks
One row per trajectory of a Qwen3 coding agent working on an MBPP+ task inside a tiny repository
(spec, empty solution.py, three visible tests; hidden EvalPlus tests grade generality). Three
experiment families (experiment column):
experiment
what is planted
how a positive is certified
native_hacking
nothing; the agent may edit/skip tests or hardcode
programmatic: test/runner diff, skip/exit, hardcode + hidden-test… See the full description on the dataset page: https://huggingface.co/datasets/Narmeen07/cot-faithfulness-agentic-traces.repro-why-agentic-theorem-prover-works-a-statistical-provability-theory-of-mathematica-traces
Agent traces
Agent sessions published from a Trackio Logbook.
grug-agentic-eval-v2-tracesagentic-coding-traces
Agentic Coding Mooncake Traces
Synthetic agentic coding benchmark datasets in Mooncake trace (JSONL) format,
generated with AIPerf 0.9.0 for LLM inference benchmarking.
Designed for use with InferenceX via the
agentic-replay scenario-type and aiperf_adapter.py.
Files
File
Sessions
Turns
max_prompt_tokens
Seed
64k/dataset.jsonl
1,000
18,595
65,536
42
128k/dataset.jsonl
1,000
16,957
131,072
42
Format
Each line is a Mooncake trace… See the full description on the dataset page: https://huggingface.co/datasets/thangquang09/agentic-coding-traces.negotiation-tracesmimo-v25-agentic-reasoning-traces
Mimo V2.5 Agentic Reasoning Traces
Repository: OusiaResearch/mimo-v25-agentic-reasoning-traces
Paper: ( forthcoming )
License: CC BY-NC 4.0
Overview
A 14,726-row corpora of agentic reasoning traces generated via Xiaomi Mimo V2.5 Pro, structured for fine-tuning and evaluating language models on consequential, real-world decision-making tasks.
The corpus has two components:
Component
Rows
Description
mimo_v25_reasoning_traces.jsonl
9,287
General reasoning traces… See the full description on the dataset page: https://huggingface.co/datasets/asakri/mimo-v25-agentic-reasoning-traces.