SurajSingh's picture
Create qa_gen.py
c1faeac verified
import os
from pathlib import Path
import numpy as np
import torch
from transformers import GPT2LMHeadModel, GPT2Tokenizer
current_path = os.path.dirname(os.path.abspath(__file__))
tokenizer_path = os.path.join(current_path, "gpt_tokenizer")
model_path = os.path.join(current_path, "gpt2_qa_model")
tokenizer = GPT2Tokenizer.from_pretrained(tokenizer_path) # also try gpt2-medium
model = GPT2LMHeadModel.from_pretrained(model_path)
def generate_text(sequence, max_new_tokens):
ids = tokenizer.encode(f'{sequence}', return_tensors='pt')
input_length = ids.size(1)
max_length = input_length + max_new_tokens
final_outputs = model.generate(
ids,
do_sample=True,
max_length=max_length,
pad_token_id=model.config.eos_token_id
)
return tokenizer.decode(final_outputs[0], skip_special_tokens=True)
def question_awnser(prompt: str):
result = generate_text("Question: " + prompt + "Answer: ", 35).split('Answer: ')[1]
try:
result = result.split('.')[0] + '.'
except Exception as e:
print(e)
return result