Model Name
model_name = "google/flan-t5-base"
Parameter for AutoTokenizer
pad_to_max_length_GD = False
max_length_GD = 64
Parameter for DataCollator
padding_in_datacollator = 'longest'
Training Arguments
batch_size = 4
lr_rate = 1e-5
num_epoch = 20
weight_decay = 0.01
fp16 = False # Make true if GPU is there
predict_with_generate=True
eval_steps = 1000
save_steps = 1000
load_best_model_at_end=True
Generating output
temperature = 1.5
max_len = 256
bum_beam = 50