CoolFace
Modelpublic

buddhist-nlp/nllb-200-1.3B-tib2eng

sourceHugging Faceupdated 4y agoView on Hugging Face
0likes19downloads
Model Card
Training setup
Datasethttps://huggingface.co/datasets/buddhist-nlp/tibengbitext
Num train steps10000
Max seq len256
Batch size512
Total data points seen5.1 mil
Total tokens seen450 mil

This is the best scoring checkpoint on the validation set at step 4400 / 10000.

SplitMetricScore
TrainBLEU36.5
ValBLEU27.1
TestBLEU27.1
Trainchrf++52.8
Valchrf++45.9
Testchrf++45.6