datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
debugtest3
DebugTestkimi-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.super-debug-v3
super-debug-v3
super-debug-v3 is a synthetic dataset of grounded software-debugging trajectories generated with hen, a stateful long-horizon AI coding agent for C/C++ projects.
This is the third version of super-debug. Compared with super-debug-v2, which focused on SimpleC/compiler debugging runs, v3 includes synthesized data across the newer hen/Projects project set:
clcalc
math3d
mini2d_tilegame
ocr8
poseblend
rigid2d
sgps
simplec
tinyvm
The default config is the… See the full description on the dataset page: https://huggingface.co/datasets/georvn7/super-debug-v3.opencode_seed2.1_video_badcase_debug_round_00agent-cwm-rubrics-debug
agent-cwm rubric library + P/R debug bundle (large split: 27 mine / 37 held-out)
Layout
library/err__*.md — 139 error rubrics (frontmatter exception_class: = the class each commits to)
library/perf_rubrics/ — 209 performance rubrics (P1 skeleton); perf_rubrics_gated/ = 92 that passed the causal gate (own patch improved own source program above measured noise; gate_manifest.json has the strict list)
library/runtime_rubrics/ — 257 runtime-cost rubrics (not part of… See the full description on the dataset page: https://huggingface.co/datasets/EdwardoSunny/agent-cwm-rubrics-debug.gpt-5.6-sol-coding-and-debugging-traces
GPT-5.6 Sol Coding & Debugging Traces
Verified software-engineering, independent model-judging, seed-authoring,
defensive-security, and training-harness trajectories from
GPT-5.6 Sol (gpt-5.6-sol) running through the Codex CLI as an
autonomous coding agent. Sessions show the observable development loop:
inspecting repositories, reproducing failures, explaining evidence, editing
files, running compilers and test suites, correcting mistakes, and verifying
the completed result.… See the full description on the dataset page: https://huggingface.co/datasets/greghavens/gpt-5.6-sol-coding-and-debugging-traces.behavior-1k-2025-challenge-demos-debugThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "R1Pro",
"total_episodes": 10000,
"total_frames": 119094660,
"total_tasks": 50,
"total_videos": 90000,
"chunks_size": 10000,
"fps": 30,
"splits": {
"train": "0:10000"
},
"data_path": "data/task-{episode_chunk:04d}/episode_{episode_index:08d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/savoji/behavior-1k-2025-challenge-demos-debug.nvidia_debugdebug_trace_runsDebugBench
Dataset Summary
DebugBench is a Large Language Model (LLM) debugging benchmark introduced in the paper DebugBench: Evaluating Debugging Capability of Large Language Models. We collect code snippets from the LeetCode community and implant bugs into source data with GPT-4. The project is also open-sourced as a GitHub repository.
It consists of 4,253 instances.
It covers four major bug categories and 18 minor types.
It includes C++, Java, and Python instances.
It contains three… See the full description on the dataset page: https://huggingface.co/datasets/Rtian/DebugBench.super-debug-v2
super-debug-v2
super-debug-v2 is a synthetic dataset of grounded software-debugging trajectories generated with hen, a stateful long-horizon AI coding agent for C/C++ projects.
This is the second version of super-debug. Compared with super-debug-v1, this release is generated from three full-suite debugging runs. The previous release kept trajectories that passed only the first three validation steps; this version keeps trajectories from runs that pass the full hen/SimpleC/tests… See the full description on the dataset page: https://huggingface.co/datasets/georvn7/super-debug-v2.glm-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.debug_upload_filesfable-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.srmk_debugcached-activationsmath-bon-debugrag-retrieval-debug-trajectories
Rag Retrieval Debug Trajectories
Rights & intended use: legacy public research corpus / portfolio
artifact. Hosted frontier-model outputs are research-only inputs under
project policy (synthetic-factory#161):
intended_use: research_only, project_training_policy: blocked. Not
training data for any model-weight update. Machine-readable record:
rights.json.
Release status: The raw, uncurated payload is now published under
data/raw/. It is available for inspection and… See the full description on the dataset page: https://huggingface.co/datasets/rmems/rag-retrieval-debug-trajectories.calvin_debug_datasetllm-debugger-sft-corpus
llm-debugger SFT corpus
Training input for llm-debugger,
a model that works a failing Python test in a live pdb session and edits the
fix. This is the corpus behind the SFT checkpoint the best RL policy (v90) was
trained from.
Contents
path
what
corpus/train_sft.jsonl
312 training rows, native tool-call format
corpus/val_sft.jsonl
35 validation rows
corpus/build_meta.json
row counts and SHA-256 per split, counted at publish
corpus/rows.jsonl
the… See the full description on the dataset page: https://huggingface.co/datasets/moofeez/llm-debugger-sft-corpus.LoVR-benchmarkThis repository hosts the LoVR benchmark dataset, designed for research on long video–text retrieval.The original video data is sourced from LongVideoBench. We sincerely thank the authors for their outstanding work.
Building upon LongVideoBench, we curate high-quality captions and semantic annotations tailored for long-form video understanding and retrieval tasks.
The corresponding GitHub repository for this project is available at:👉 https://github.com/TechNomad-ds/LoVR-benchmark
Below we… See the full description on the dataset page: https://huggingface.co/datasets/debugger123/LoVR-benchmark.critichack-fr3-middle-drawer-debugkimi-k3-coding-and-debugging-traces
Kimi K3 Coding, Tool Use & Instruction Following Traces
697 TRAJECTORIES · 4,890 TRAINING ROWS · 3 MB PARQUET · 89 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/jiajiale9/kimi-k3-coding-and-debugging-traces.lichess_elo_binned_debugrllm-tb-v2-debug-fireworks-batchcode-debug-outputscua_debugger_traj
CUA Debugger Trajectories
204 failed computer-use agent (CUA) trajectories on OSWorld, each with a human root-cause annotation.
Three agents were run on OSWorld (Ubuntu desktop, screenshot-only observation, pyautogui execution at 1920×1080). Every trajectory in this dataset is a failure (no task reached evaluator score 1.0). For each trajectory, a human annotator identified the root error step — the earliest step responsible for the failure — and labeled it with an… See the full description on the dataset page: https://huggingface.co/datasets/CyT1ng/cua_debugger_traj.lib90-rotate-debugged-90This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "panda",
"total_episodes": 90,
"total_frames": 18110,
"total_tasks": 74,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 10,
"splits": {
"train": "0:90"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/windfromthenorth/lib90-rotate-debugged-90.generations-21-DEBUG-qwen3-8b-simnpo-gentle-igm-10b-target-100-localtrain-checkpoint-1