https://huggingface.co/seanghay/whisper-small-khmer
conda create -n whisper transformers
conda activate whisper
pip install -r requirements.txt
Login with huggingface.co
huggingface-cli login
Start
python3 seq2seq.py
Adjust these if got CUDA error: out of memory!
per_device_train_batch_size=64, # decrease by 2x
per_device_eval_batch_size=32, # decrease by 2x
gradient_accumulation_steps=1, # increase 2x
# effective_batch_size = gradient_accumulation_steps x per_device_train_batch_size