config.yaml

output_dir: ./training
batch_size: 16

saved_gcp: True
gcp_path: gs://kaggle-vbdi-test/nlp-text-summarization/shorten-sentence

load_pretrained: False
gcp_pretrained_path: gs://kaggle-vbdi-test/nlp-text-summarization/15epochs/checkpoint-65000

# Tokenizer setup 
encoder_max_length: 256
decoder_max_length: 32
number_sentences_original: 10

# Setting up decoder config
max_length: 32
early_stopping : True
no_repeat_ngram_size : 3
length_penalty : 2.0
num_beams : 4

# Setting up trainer configs
predict_with_generate : True
do_train : True
do_eval: True
logging_steps : 200  
save_steps: 5000 
eval_steps: 7500 
warmup_steps: 3000  
num_train_epochs: 10
overwrite_output_dir : True
save_total_limit: 30
fp16: True