datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Repaired_videosex-repairci-repair-bench
CI-REPAIR-BENCH
Overview
CI-REPAIR-BENCH is a benchmark dataset for research on Continuous Integration (CI) failures and automated repair in Python repositories.
The dataset contains 567 CI failure instances collected from 105 real-world GitHub repositories, all written in Python.Each instance captures a CI workflow failure, its logs, the corresponding code diff, and repository-level metadata.
Dataset Statistics
Programming language: Python
Number… See the full description on the dataset page: https://huggingface.co/datasets/ci-benchmark-user/ci-repair-bench.mlqa_repairedThis is a repaired version of https://huggingface.co/datasets/facebook/mlqa made compatible with datasets>=4.X (no arbitrary code execution).
data-pipeline-repair-trajectories
Data Pipeline Repair 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/data-pipeline-repair-trajectories.SWEUniverse-Repaired-Indist-full-not-SWE-bench-pro-matched
VmaxRL/SWEUniverse-Repaired-Indist-full-not-SWE-bench-pro-matched
This dataset contains a 350-row subset selected from the Indist SWEUniverse training rows.
Selection policy: three-way repo overlap with Bugpilot and LM-Modify, deduped by repo plus introduction patch, then balanced round-robin across overlapping repos.
Rows: 350
Selected repos: 19
Deduped overlap capacity: 468
Source dataset: VmaxRL/SWEUniverse-Repaired-Indist-full-not-SWE-bench-pro-matched
repairllama-datasets
RepairLLaMA - Datasets
Contains the processed fine-tuning datasets for RepairLLaMA.
Instructions to explore the dataset
To load the dataset, you must define which revision (i.e., which input/output representation pair) you want to load.
from datasets import load_dataset
# Load ir1xor1
dataset = load_dataset("ASSERT-KTH/repairllama-datasets", "ir1xor1")
# Load irXxorY
dataset = load_dataset("ASSERT-KTH/repairllama-datasets", "irXxorY")
Citation
If you use… See the full description on the dataset page: https://huggingface.co/datasets/ASSERT-KTH/repairllama-datasets.VeriLoop-Structural-Repair-Verified
VLR-StructuralRepair v1.0.0 — non-regressive repair of real semantic defects
Evidence-convergent supervision for function-level semantic repair under a
hidden set of protected obligations. A candidate is positive only when it
preserves every already-satisfied obligation and strictly repairs at least
one. Aggregate improvement that breaks a protected obligation is a negative,
however far the total failure count drops.
The previous generation of this dataset… See the full description on the dataset page: https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-Structural-Repair-Verified.Minecraft-GLB2Schem-RepairPairs-v1
unfundedResearcher/Minecraft-GLB2Schem-RepairPairs-v1
Paired (generated input, ground-truth target) Minecraft schematics for training a
model that turns an approximate voxelisation into a real build.
What a sample is
Each sample is three files inside a WebDataset TAR shard:
File
Meaning
<id>.input.schem
GENERATED. Produced by voxelising the source .glb. Approximate and noisy.
<id>.target.schem
GROUND TRUTH. The original schematic, copied byte-for-byte… See the full description on the dataset page: https://huggingface.co/datasets/unfundedResearcher/Minecraft-GLB2Schem-RepairPairs-v1.SWESwiss-Repair-RL-SWEGym-SWESmith-12K
Overview
RL dataset for training SWE-Swiss models on the repair task. The prompts are based on issues from SWE-Gym and SWE-smith. To create a challenging task, the code content in each prompt consists of two components: "oracle" files, which are the ground-truth files requiring a patch, and "distractor" files, which are plausible but incorrect files predicted by an LLM.
Citation
@misc{SWESwiss2025,
title = {SWE-Swiss: A Multi-Task Fine-Tuning and RL Recipe for… See the full description on the dataset page: https://huggingface.co/datasets/SWE-Swiss/SWESwiss-Repair-RL-SWEGym-SWESmith-12K.indicxnli_repaired
IndicXNLI (Repaired)
This is a repaired version of the Divyanshu/indicxnli dataset, converted to parquet format for compatibility with HuggingFace datasets 4.x+.
Why this exists
The original dataset uses a Python loading script (indicxnli.py) which is no longer supported in HuggingFace datasets 4.x. This version converts the data to native parquet format.
Original Dataset
Paper: IndicXNLI: Evaluating Multilingual Inference for Indian Languages
Original Repo:… See the full description on the dataset page: https://huggingface.co/datasets/AdaMLLab/indicxnli_repaired.OpenCOCO-I2T-Repack-Tiny
OpenCOCO-I2T-Repack-Tiny
OpenCOCO-I2T-Repack-Tiny is a compact image-to-text / image-text-to-text captioning dataset containing 158,958 image samples sourced from the COCO dataset and repackaged into a lightweight format suitable for vision-language model (VLM) fine-tuning. The dataset contains synthesized responses generated using a custom Qwen3.5 multimodal captioning pipeline. The input images undergo lossless image compression to significantly reduce the overall storage… See the full description on the dataset page: https://huggingface.co/datasets/prithivMLmods/OpenCOCO-I2T-Repack-Tiny.SWEUniverse-Repaired-Bugpilot
SWEUniverse Repaired Bugpilot
Reformatted repaired dataset from VmaxRL/SWEUniverse-Repaired-Bugpilot.
Rows: 804
Columns are exactly:
Task id
initial p2p
new p2p
initial f2p
new f2p
default test command
new test command
test parser
image name
patch
problem statement
repo
initial f2p/initial p2p are the source labels before repair; new f2p/new p2p are the repaired labels used for evaluation.
ReAPR-Automatic-Program-Repair-via-Retrieval-Augmented-Large-Language-ModelsThis is the Retrieval dataset used in the paper "ReAPR: Automatic Program Repair via Retrieval-Augmented Large Language Models"
lca-ci-builds-repair
🏟️ Long Code Arena (CI builds repair)
This is the benchmark for CI builds repair task as part of the
🏟️ Long Code Arena benchmark.
🛠️ Task. Given the logs of a failed GitHub Actions workflow and the corresponding repository snapshot,
repair the repository contents in order to make the workflow pass.
All the data is collected from repositories published under permissive licenses (MIT, Apache-2.0, BSD-3-Clause, and BSD-2-Clause). The datapoints can be removed upon request.
To… See the full description on the dataset page: https://huggingface.co/datasets/JetBrains-Research/lca-ci-builds-repair.moss-voice-identity-repairs
MOSS voice-acting v2 -- repaired takes
For each voice profile, every take whose ECAPA speaker similarity to the voice's reference fell
below 0.40, regenerated with that voice's identity LoRA (see
laion/moss-voice-identity-loras) merged at scale 1.0 on top of the identical condition
adapters at the identical lambdas.
Nothing here replaces anything. The original takes are untouched and remain part of the
corpus; low-similarity takes are kept deliberately, because they are useful… See the full description on the dataset page: https://huggingface.co/datasets/laion/moss-voice-identity-repairs.appliancedb-error-codes-repair-database
ApplianceDB: Home Appliance Error Codes & Ranked Repairs
Full dataset: appliancedb.dataengineered.io · $99 one-time (Repair Intelligence Snapshot: commercial licence + SQLite and Parquet builds; the same rows as this sample) → Buy on Stripe · the same sample on Kaggle
Relational database mapping 438 home-appliance error codes across 13 brands and 26 (brand, appliance-type) pairs to 288 ranked repair procedures with DIY difficulty tiers. Every code is identified by its… See the full description on the dataset page: https://huggingface.co/datasets/Ichlibitiche/appliancedb-error-codes-repair-database.SWEUniverse-Repaired-Indist
SWEUniverse Repaired Indist
Cleaned repaired Indist dataset. Rows with empty requested fields were removed.
Rows before cleaning: 846
Rows after cleaning: 796
Dropped rows: 50
Columns are exactly:
Task id
initial p2p
new p2p
initial f2p
new f2p
default test command
new test command
test parser
image name
patch
problem statement
repo
SWEUniverse-Repaired-LM-Modify
SWEUniverse Repaired LM-Modify
Reformatted repaired dataset from VmaxRL/bugpilot-bugintro-lm-modify-gpt55-1k_repaired.
Rows: 706
Columns are exactly:
Task id
initial p2p
new p2p
initial f2p
new f2p
default test command
new test command
test parser
image name
patch
problem statement
repo
initial f2p/initial p2p are the source labels before repair; new f2p/new p2p are the repaired labels used for evaluation.
grug-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.polaris-53k-repaired
POLARIS-53K, label-repaired
49,289 of the 53,291 rows in
POLARIS-Project/Polaris-Dataset-53K,
with 4,580 stored answers corrected and 4,002 rows removed as unrepairable.
Measurements on the source set put its bad-label rate at roughly 15.9%
[14.3, 17.6] (two independent detectors agreeing on a 2,000-row sample).
Mislabelled rows are not uniformly distributed: they concentrate in the problems
models fail, which is exactly where a difficulty-calibration pipeline looks.… See the full description on the dataset page: https://huggingface.co/datasets/joanvelja/polaris-53k-repaired.mechanicdb-obd2-repair-sample
🔧 MechanicDB — OBD-II Diagnostic & Repair Database (Free Sample)
Full dataset: mechanicdb.dataengineered.io · $49 Standard (SAE) · $149 OEM Complete, one-time → Buy Standard · Buy OEM Complete · the same sample on Kaggle
The free developer sample of MechanicDB: an automotive dataset mapping OBD-II
Diagnostic Trouble Codes (DTCs) to ranked repair procedures with DIY
difficulty ratings, aftermarket parts-cost ranges (USD), labor-hour estimates,
and step-by-step instructions.
90… See the full description on the dataset page: https://huggingface.co/datasets/Ichlibitiche/mechanicdb-obd2-repair-sample.tabrepair-science-repair-under-shift
TabRepair Science: Repair Under Shift
TabRepair Science is a finite authored benchmark for a deceptively hard
question: does better tabular cell repair produce better downstream models
under distribution shift?
The 3,648-row pilot spans three structural generator families, missingness and
present-value contamination, four test regimes, eight repair representations,
and five downstream learners. A separate eight-world sensitivity layer tests a
damage-aware v2 candidate without… See the full description on the dataset page: https://huggingface.co/datasets/haidang2405/tabrepair-science-repair-under-shift.SWEUniverse-Repaired-Bugpilot-350
SWEUniverse-Repaired-Bugpilot-350
A deterministic 350-task subset in the swe_fix_training_row_v1 format introduced in PR #1942.
Schema: swe_fix_training_row_v1
Source dataset: VmaxRL/SWEUniverse-Repaired-Bugpilot
Rows: 350
Split: train
Task variant: generated_bug_intro
Source kind: swesmith
Generator id: bugpilot
The subset was sampled from validated accepted tasks and preserves the validated repository distribution as closely as possible. pass_to_pass excludes any labels also… See the full description on the dataset page: https://huggingface.co/datasets/VmaxRL/SWEUniverse-Repaired-Bugpilot-350.SWEUniverse-Repaired-LM-Modify-350
SWEUniverse-Repaired-LM-Modify-350
A deterministic 350-task subset in the swe_fix_training_row_v1 format introduced in PR #1942.
Schema: swe_fix_training_row_v1
Source dataset: VmaxRL/SWEUniverse-Repaired-LM-Modify
Rows: 350
Split: train
Task variant: generated_bug_intro
Source kind: swesmith
Generator id: lm_modify
The subset was sampled from validated accepted tasks and preserves the validated repository distribution as closely as possible. pass_to_pass excludes any labels also… See the full description on the dataset page: https://huggingface.co/datasets/VmaxRL/SWEUniverse-Repaired-LM-Modify-350.satd-repayment-context
SATD Repayment Context Dataset
Extends the SATD Repayment replication package (Python: 58,722 rows,
Java: 97,347 rows) with per-SATD repository context, all anchored at
parent(deleted_in_commit) (the commit right before the fix), so no information
from the repayment itself leaks into the context. Total package size: ~1.56 GB.
Contents
data/
python_final.parquet -- main SATD table (Python), 58,722 rows, 130 MB
java_final.parquet -- main SATD table… See the full description on the dataset page: https://huggingface.co/datasets/ngducloc1112002/satd-repayment-context.dfm11-toolace-native-tool-use-repaired
dfm11-toolace-native-tool-use-repaired
ToolACE conversations with declared-name parsing and complete parallel result binding.
This is a DFM11 replacement for schneiderkamplab/dfm10-toolace-native-tool-use. All rows pass exhaustive structural validation. See metadata/manifest.json.
semantic-repair-routing
semantic-repair-routing
The 84,819 supervised pairs that trained
SemanticRepair-270M:
a message somebody actually wrote, and the requests inside it restated
plainly, one per line.
It teaches one narrow thing. An embedding router compares a question with
the description of every capability it can reach. People do not write the
way capabilities are described — they hedge, they apologise, they ask two
things in one breath, they name what they do not want. This data pairs
the first… See the full description on the dataset page: https://huggingface.co/datasets/Gramscii-IT/semantic-repair-routing.indist-tool-v0-pool-v2-gpt55-1k_issue_rewritten_prompt-v3-v5_swesmith_metadata_repairedcode_repair