File size: 983 Bytes
0df4b7b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 |
MAX_INPUT_LENGTH = 256
MAX_TARGET_LENGTH = 128
def preprocess_function(examples):
"""
Preprocess entries of the given dataset (should be used with a `map` function)
Params:
examples (Dataset): dataset to be preprocessed
Returns:
model_inputs (BatchEncoding): tokenized dataset entries
"""
inputs, targets = [], []
for i in range(len(examples['question'])):
inputs.append(f"Antwort: {examples['provided_answer'][i]} Lösung: {examples['reference_answer'][i]} Frage: {examples['question'][i]}")
targets.append(f"{examples['score'][i]} Feedback: {examples['answer_feedback'][i]}")
# apply tokenization to inputs and labels
model_inputs = tokenizer(inputs, max_length=MAX_INPUT_LENGTH, padding='max_length', truncation=True)
labels = tokenizer(text_target=targets, max_length=MAX_TARGET_LENGTH, padding='max_length', truncation=True)
model_inputs['labels'] = labels['input_ids']
return model_inputs
|