def predict(self,
questions,
search_by='answer',
topk=5,
answer_only=False):
embedding = self.qa_embed.predict(
questions=questions, dataset=False).eval(session=self.embed_sess)
if answer_only:
topk_answer = self.faiss_topk.predict(embedding, search_by, topk,
answer_only)
else:
topk_question, topk_answer = self.faiss_topk.predict(
embedding, search_by, topk, answer_only)
gpt2_input = self._get_gpt2_inputs(questions, topk_question,
topk_answer)
raw_output = gpt2.generate(self.sess,
prefix=gpt2_input,
return_as_list=True,
include_prefix=False)
original_line = '`QUESTION: %s `ANSWER: ' % questions
output_list = []
for output_ind, output_chunk in enumerate(
raw_output[0].split(original_line)):
if output_ind == 0:
pass
else:
output_list.append(output_chunk.split('`QUESTION')[0])
# clipped_output = raw_output[0].split(
# '`QUESTION')[1].split('`ANSWER:')[1]
return output_list
import tensorflow.compat.v1 as tf
from gpt_2 import gpt2
tf.disable_eager_execution()
csv_path = 'data/GPT2_data_FFNN.csv'
sess = gpt2.start_tf_sess()
gpt2.load_gpt2(sess)
gpt2.generate(
sess, prefix="`QUESTION: What is the best treatment for the flu? `ANSWER:")