deberta-vie-conv / deberta_train.log
hieule's picture
deberta and replace linear with conv depth wise
11a40e3
raw
history blame contribute delete
514 Bytes
DATASET: corpus_v1_10000000 | Model: deberta-vie-small
>>> Train Epochs: 1 | Losses: 2.423206581025543 | lr: 0.00013761917313292747
>>> Eval Epochs: 1 | Losses: 1.8951771660055028 | Perplexity: 6.673118307905794
>>> Train Epochs: 2 | Losses: 1.8172417448716809 | lr: 6.880958656646373e-05
>>> Eval Epochs: 2 | Losses: 1.6318174320577783 | Perplexity: 5.106666322469493
>>> Train Epochs: 3 | Losses: 1.6091945579320812 | lr: 0.0
>>> Eval Epochs: 3 | Losses: 1.487273468930144 | Perplexity: 4.307366906292279