| Usage: | |
| ``` | |
| import nlp2 | |
| import json | |
| from datasets import load_dataset | |
| from transformers import AutoTokenizer, AutoModelForSeq2SeqLM | |
| from asrp.code2voice_model.hubert import hifigan_hubert_layer6_code100 | |
| import IPython.display as ipd | |
| tokenizer = AutoTokenizer.from_pretrained("Oscarshih/long-t5-base-SQA-15ep") | |
| model = AutoModelForSeq2SeqLM.from_pretrained("Oscarshih/long-t5-base-SQA-15ep") | |
| dataset = load_dataset("voidful/NMSQA-CODE") | |
| cs = hifigan_hubert_layer6_code100() | |
| qa_item = dataset['dev'][0] | |
| question_unit = json.loads(qa_item['hubert_100_question_unit'])[0]["merged_code"] | |
| context_unit = json.loads(qa_item['hubert_100_context_unit'])[0]["merged_code"] | |
| answer_unit = json.loads(qa_item['hubert_100_answer_unit'])[0]["merged_code"] | |
| # groundtruth answer | |
| ipd.Audio(data=cs(answer_unit), autoplay=False, rate=cs.sample_rate) | |
| # predict answer | |
| inputs = tokenizer("".join([f"v_tok_{i}" for i in question_unit]) + "".join([f"v_tok_{i}" for i in context_unit]), return_tensors="pt") | |
| code = tokenizer.batch_decode(model.generate(**inputs,max_length=1024))[0] | |
| code = [int(i) for i in code.replace("</s>","").replace("<s>","").split("v_tok_")[1:]] | |
| ipd.Audio(data=cs(code), autoplay=False, rate=cs.sample_rate) | |
| ``` |