KrazyKitty/Data-Lab
Updated • 56
KRAZYKITTY // DATA PROVENANCE & EVALUATION
Transparent corpus construction, source governance, filtering research, benchmark documentation, and contamination-aware evaluation.
Note Documentation-only provenance, consent, quality, preprocessing, and evaluation lab; no dataset corpus is published.
Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research
Paper • 2402.00159 • Published • 66Note Open corpus construction and documented data-curation research.
The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale
Paper • 2406.17557 • Published • 105Note Large-scale web-corpus filtering, deduplication, and ablation study.
DataComp-LM: In search of the next generation of training sets for language models
Paper • 2406.11794 • Published • 54Note DataComp-LM benchmark and experiments for language-model dataset curation.
allenai/dolma
Updated • 2.89k • 1.11kNote Open pretraining corpus; review source-specific terms and current card before use.
HuggingFaceFW/fineweb
Viewer • Updated • 52.5B • 363k • 3.73kNote Documented web-corpus repository; pin a revision and review source rights before use.
openai/gsm8k
Benchmark • Updated • 17.6k • 1.08M • 2.03kNote Reasoning evaluation dataset; keep evaluation content out of training mixtures.
Idavidrein/gpqa
Benchmark • Updated • 1.25k • 98.5k • 634Note Gated expert QA benchmark. Preserve the current restriction against revealing examples.