Kelvinhjk's picture
Update app.py
b91137f
raw
history blame
2.27 kB
import os
import streamlit as st
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import FAISS
from transformers import TFAutoModelForQuestionAnswering, AutoTokenizer, pipeline
os.environ["OPENAI_API_KEY"] = "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"
# Read data
with open("./data/full_context.txt", "r") as file1:
doc = file1.read()
# Splitting up the text into smaller chunks for indexing
text_splitter = CharacterTextSplitter(
separator = "\n",
chunk_size = 1000,
chunk_overlap = 200, #striding over the text
length_function = len,
)
texts = text_splitter.split_text(doc)
# Download embeddings from OpenAI
embeddings = OpenAIEmbeddings()
docsearch = FAISS.from_texts(texts, embeddings)
# Load model
model_path = "/content/drive/MyDrive/Colab_Notebooks/COS30081_NLP/D_HD_Task/models/roberta_model"
model = TFAutoModelForQuestionAnswering.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained('deepset/roberta-base-squad2')
# Initialize Transformer pipeline with our own model and tokenizer
question_answerer = pipeline("question-answering", model=model, tokenizer=tokenizer)
def findHighestScore(question):
docs_found = docsearch.similarity_search(question)
doc_score = 0.5
doc_answer = ''
for doc in docs_found:
doc_result = question_answerer(question=question, context = doc.page_content)
if doc_result['score'] > doc_score:
doc_score = doc_result['score']
doc_answer = doc_result['answer']
return doc_answer, doc_score
def QnAfunction(question):
answer1, score1 = findHighestScore(question)
if answer1 != '':
return answer1, score1
# print("Answer: ", answer1)
# print("Score: ", score1)
else:
return "No Answer found. Please ask question related to Bachelor of Computer Science program at Swinburne.", 0
# print("No Answer found. Please ask question related to Bachelor of Computer Science program at Swinburne.")
text = st.text_area("Ask any question about the Bachelor of Computer Science program at Swinburne: ")
if text:
ans, score = QnAfunction(text)
st.json(ans)