samarthramesh/gaperon-distill
gaperon-distill Reformatted, lean parquet build of the Gaperon mmBERT quality-distillation training data. One config per language (english, hindi, tamil), each with train / validation / test splits preserved exactly from the original make_splits partition (seed 42, 80/10/10). Built for fast loading on a cluster with no persistent storage. from datasets import load_dataset ds = load_dataset("samarthramesh/gaperon-distill", "hindi", split="train") Columns… See the full description on the dataset page: https://huggingface.co/datasets/samarthramesh/gaperon-distill.
gaperon-distill
Reformatted, lean parquet build of the Gaperon mmBERT quality-distillation training data. One config per language (english, hindi, tamil), each with train / validation / test splits preserved exactly from the original make_splits partition (seed 42, 80/10/10). Built for fast loading on a cluster with no persistent storage.
from datasets import load_dataset
ds = load_dataset("samarthramesh/gaperon-distill", "hindi", split="train")Columns
All records are fully labelled (any record with an unknown/missing dimension was dropped, matching the original training filter).