predictor-part2.py

# predictor.py

def predict(sample, metadata):
    indexed_tokens = tokenizer.encode(sample["text"])
    output = sample_sequence(model, metadata['num_words'], indexed_tokens, device=metadata['device'])
    return tokenizer.decode(
        output[0, 0:].tolist(), clean_up_tokenization_spaces=True, skip_special_tokens=True
    )