awong-dev/whisper-large-v3-cantonese

File-tuned on common voice 25 yue dataset. Full run with punctuation removed before CER calculation. LR 3e-6, 3 epochs, cosine schedule

Trained on common voice 22 yue using 5% test, .5% eval and lr 3e-6 (TensorBoard runs)

Trained on common voice 22 yue using 5% test, .5% eval and lr 3e-6

Trained on common voice 22 yue using train.tsv as the validation dataset and the remainder of validated.tsv for training set (TensorBoard runs)

Trained on common voice 22 yue using train.tsv as the validation dataset and the remainder of validated.tsv for training set

initial commit