Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -138,3 +138,95 @@
|
|
| 138 |
|
| 139 |
|
| 140 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 138 |
|
| 139 |
|
| 140 |
|
| 141 |
+
import streamlit as st
|
| 142 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
| 143 |
+
from langchain.vectorstores import FAISS
|
| 144 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
| 145 |
+
from langchain.document_loaders import PyPDFLoader
|
| 146 |
+
from langchain.chains import RetrievalQA
|
| 147 |
+
from langchain.llms import HuggingFaceHub
|
| 148 |
+
import tempfile
|
| 149 |
+
import os
|
| 150 |
+
|
| 151 |
+
# Constants
|
| 152 |
+
EMBEDDING_MODEL_NAME = "BAAI/bge-base-en-v1.5"
|
| 153 |
+
LLM_MODEL_REPO = "mistralai/Mistral-7B-Instruct-v0.1"
|
| 154 |
+
CHUNK_SIZE = 500
|
| 155 |
+
CHUNK_OVERLAP = 300
|
| 156 |
+
|
| 157 |
+
# Load and split documents
|
| 158 |
+
def load_and_split_pdf(pdf_file):
|
| 159 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp_file:
|
| 160 |
+
tmp_file.write(pdf_file.read())
|
| 161 |
+
tmp_file_path = tmp_file.name
|
| 162 |
+
|
| 163 |
+
loader = PyPDFLoader(tmp_file_path)
|
| 164 |
+
documents = loader.load()
|
| 165 |
+
splitter = RecursiveCharacterTextSplitter(chunk_size=CHUNK_SIZE, chunk_overlap=CHUNK_OVERLAP)
|
| 166 |
+
chunks = splitter.split_documents(documents)
|
| 167 |
+
return chunks
|
| 168 |
+
|
| 169 |
+
# Create FAISS vectorstore
|
| 170 |
+
def build_vectorstore(chunks):
|
| 171 |
+
embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL_NAME)
|
| 172 |
+
db = FAISS.from_documents(chunks, embedding=embeddings)
|
| 173 |
+
return db
|
| 174 |
+
|
| 175 |
+
# Initialize LLM from Hugging Face Hub
|
| 176 |
+
def get_llm():
|
| 177 |
+
return HuggingFaceHub(
|
| 178 |
+
repo_id=LLM_MODEL_REPO,
|
| 179 |
+
model_kwargs={"temperature": 0.3, "max_new_tokens": 512, "top_k": 10}
|
| 180 |
+
)
|
| 181 |
+
|
| 182 |
+
# Custom prompt for better accuracy
|
| 183 |
+
CUSTOM_PROMPT = """
|
| 184 |
+
You are a professional resume chatbot. Use the context below to accurately and concisely answer the user's question. If the information is not available in the context, say "Not found in the document.".
|
| 185 |
+
|
| 186 |
+
Context:
|
| 187 |
+
{context}
|
| 188 |
+
|
| 189 |
+
Question:
|
| 190 |
+
{question}
|
| 191 |
+
|
| 192 |
+
Answer:
|
| 193 |
+
"""
|
| 194 |
+
|
| 195 |
+
# Build QA chain
|
| 196 |
+
def build_qa_chain(vectorstore):
|
| 197 |
+
return RetrievalQA.from_chain_type(
|
| 198 |
+
llm=get_llm(),
|
| 199 |
+
retriever=vectorstore.as_retriever(),
|
| 200 |
+
chain_type="stuff",
|
| 201 |
+
chain_type_kwargs={
|
| 202 |
+
"prompt": CUSTOM_PROMPT
|
| 203 |
+
}
|
| 204 |
+
)
|
| 205 |
+
|
| 206 |
+
# Streamlit UI
|
| 207 |
+
def main():
|
| 208 |
+
st.set_page_config(page_title="Resume Q&A Bot", layout="wide")
|
| 209 |
+
st.title("Resume Chatbot - Ask Anything About the Uploaded PDF")
|
| 210 |
+
|
| 211 |
+
uploaded_file = st.file_uploader("Upload your resume (PDF)", type="pdf")
|
| 212 |
+
|
| 213 |
+
if uploaded_file is not None:
|
| 214 |
+
st.success("PDF uploaded successfully!")
|
| 215 |
+
with st.spinner("Processing document and creating knowledge base..."):
|
| 216 |
+
chunks = load_and_split_pdf(uploaded_file)
|
| 217 |
+
vectorstore = build_vectorstore(chunks)
|
| 218 |
+
qa_chain = build_qa_chain(vectorstore)
|
| 219 |
+
|
| 220 |
+
st.success("Knowledge base ready! Ask your question below:")
|
| 221 |
+
|
| 222 |
+
question = st.text_input("Your Question:")
|
| 223 |
+
|
| 224 |
+
if question:
|
| 225 |
+
with st.spinner("Generating answer..."):
|
| 226 |
+
response = qa_chain.run(question)
|
| 227 |
+
st.markdown(f"**Answer:** {response}")
|
| 228 |
+
|
| 229 |
+
if __name__ == '__main__':
|
| 230 |
+
main()
|
| 231 |
+
|
| 232 |
+
|