datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
slm-parameter-audit
SLM card-vs-artifact parameter audit
An autonomous audit of small-language-model repos on the Hugging Face Hub. For each
in-scope model (independent builders training very small models from scratch, roughly
0.5M–500M parameters), the parameter count stated in the model card is compared against
the actual artifact: the safetensors header, config.json, and the training script where
present. A mismatch is recorded when the card's number does not match the artifact's
real parameter… See the full description on the dataset page: https://huggingface.co/datasets/Compactbot/slm-parameter-audit.Gargantua-R1-Compact
Gargantua-R1 Distribution
Gargantua-R1-Compact(experimental purpose)
Gargantua-R1-Compact is a large-scale, high-quality reasoning dataset primarily designed for mathematical reasoning and STEM education. It contains approximately 6.67 million problems and solution traces, with a strong emphasis on mathematics (over 70%), as well as coverage of scientific domains, algorithmic challenges, and creative logic puzzles. The dataset is suitable for training and evaluating… See the full description on the dataset page: https://huggingface.co/datasets/prithivMLmods/Gargantua-R1-Compact.l0-essentials-compact-ablation
L0 Essentials compact ablation — raw rows
Row-level outputs from ablating the MOBIUS MMV L0 Essentials governance prompt
on six local models, three axes (tool loop / false premise / abstain chat), with
the scorer sources and the pre-registered predictions.
Status: experimental; not adversarially reviewed. The scorers had five
documented defects during the work (all fixed, rows rescored); raw outputs are
included so you can rescore with your own instrument.
Headline… See the full description on the dataset page: https://huggingface.co/datasets/moebiusT7/l0-essentials-compact-ablation.dfm14-dala-v2-gl-compact
DaLA v2 — Galician — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
807,437
807… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-gl-compact.dfm14-dala-v2-tr-compact
DaLA v2 — Turkish — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
920,794
920… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-tr-compact.dfm14-dala-v2-ar-compact
DaLA v2 — Modern Standard Arabic — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-ar-compact.dfm14-dala-v2-zh-compact
DaLA v2 — Chinese — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
1,037,788
1… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-zh-compact.Gargantua-R1-Compact
Gargantua-R1 Distribution
Gargantua-R1-Compact(experimental purpose)
Gargantua-R1-Compact is a large-scale, high-quality reasoning dataset primarily designed for mathematical reasoning and STEM education. It contains approximately 6.67 million problems and solution traces, with a strong emphasis on mathematics (over 70%), as well as coverage of scientific domains, algorithmic challenges, and creative logic puzzles. The dataset is suitable for training and evaluating… See the full description on the dataset page: https://huggingface.co/datasets/introvoyz041/Gargantua-R1-Compact.dfm14-dala-v2-ga-compact
DaLA v2 — Irish — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
824,204
824,204… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-ga-compact.dfm14-dala-v2-mk-compact
DaLA v2 — Macedonian — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
795,977
795… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-mk-compact.dfm14-dala-v2-cy-compact
DaLA v2 — Welsh — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
1,046,231
1,046… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-cy-compact.dfm14-dala-v2-vi-compact
DaLA v2 — Vietnamese — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
859,656
859… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-vi-compact.dfm14-dala-v2-ru-compact
DaLA v2 — Russian — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
823,728
823… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-ru-compact.dfm14-dala-v2-ja-compact
DaLA v2 — Japanese — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
911,381
911… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-ja-compact.dfm14-dala-v2-he-compact
DaLA v2 — Hebrew — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
1,049,799
1,049… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-he-compact.dfm14-dala-v2-eu-compact
DaLA v2 — Basque — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
887,025
887,025… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-eu-compact.dfm14-dala-v2-mt-compact
DaLA v2 — Maltese — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
807,999
807… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-mt-compact.dfm14-dala-v2-id-compact
DaLA v2 — Indonesian — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
879,440
879… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-id-compact.dfm14-dala-v2-ko-compact
DaLA v2 — Korean — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
833,536
833,536… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-ko-compact.dfm14-dala-v2-hi-compact
DaLA v2 — Hindi — DFM14 compact audit subset
One language dataset combining all selected base and additive runs. Two task views share the same accepted source records; they are not independent observations. Each independently audited clean control contributes one correct input per task; each passing pair contributes one corrupted input per task. No extra clean controls are synthesized.
Split/view
Acceptability rows
Correction rows
train_representative
1,038,773
1,038… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm14-dala-v2-hi-compact.dfm13-dala-v2-lv-compact
dfm13-dala-v2-lv-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-lv-compact.dfm13-dala-v2-pt-PT-compact
dfm13-dala-v2-pt-PT-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-pt-PT-compact.dfm13-dala-v2-en-compact
dfm13-dala-v2-en-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-en-compact.dfm13-dala-v2-lb-compact
dfm13-dala-v2-lb-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-lb-compact.dfm13-dala-v2-lt-compact
dfm13-dala-v2-lt-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-lt-compact.dfm13-dala-v2-bs-compact
dfm13-dala-v2-bs-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-bs-compact.MedConclusion-Compact
MedConclusion-Compact
MedConclusion is a large-scale dataset of 5.7M PubMed structured abstracts for biomedical conclusion generation. Each instance pairs the non-conclusion sections of an abstract with the original author-written conclusion, providing naturally occurring supervision for evidence-to-conclusion reasoning. MedConclusion also includes journal-level metadata such as biomedical category and SJR, enabling subgroup analysis across biomedical domains.
This repository… See the full description on the dataset page: https://huggingface.co/datasets/harvardairobotics/MedConclusion-Compact.dfm13-dala-v2-sr-compact
dfm13-dala-v2-sr-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-sr-compact.dfm13-dala-v2-fa-compact
dfm13-dala-v2-fa-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-fa-compact.dfm13-dala-v2-pl-compact
dfm13-dala-v2-pl-compact
One language dataset combining accepted baseline and recovery pools where available. Exact native conversations are deduplicated within each task/split; split conflicts fail closed. Full native messages, target indices and source provenance are preserved; publication_pool records origin. Task configs separate acceptability and correction; heldout views never become train. Automated compact audits are not native-speaker or producer per-edit certification.… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm13-dala-v2-pl-compact.