Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -483,26 +483,45 @@ def setup_qa(db):
|
|
| 483 |
return RetrievalQA.from_chain_type(llm=llm, retriever=retriever, return_source_documents=True)
|
| 484 |
|
| 485 |
# --- Process Answer ---
|
| 486 |
-
def process_answer(
|
| 487 |
-
|
| 488 |
-
|
|
|
|
| 489 |
|
| 490 |
-
|
| 491 |
-
|
| 492 |
-
|
| 493 |
|
| 494 |
-
|
| 495 |
-
|
| 496 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 497 |
|
| 498 |
-
{user_question}
|
| 499 |
-
"""
|
| 500 |
-
try:
|
| 501 |
-
result = qa({"query": prompt})
|
| 502 |
-
return result['result']
|
| 503 |
-
except Exception as e:
|
| 504 |
-
logging.error(f"Error generating answer: {e}")
|
| 505 |
-
return "Sorry, I couldn't generate an answer due to an internal error."
|
| 506 |
|
| 507 |
# --- UI Layout ---
|
| 508 |
with st.sidebar:
|
|
|
|
| 483 |
return RetrievalQA.from_chain_type(llm=llm, retriever=retriever, return_source_documents=True)
|
| 484 |
|
| 485 |
# --- Process Answer ---
|
| 486 |
+
def process_answer(question, full_text):
|
| 487 |
+
# STEP 1: Chunk the PDF text
|
| 488 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100)
|
| 489 |
+
docs = text_splitter.create_documents([full_text])
|
| 490 |
|
| 491 |
+
# STEP 2: Create embeddings
|
| 492 |
+
embeddings = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
|
| 493 |
+
db = Chroma.from_documents(docs, embeddings)
|
| 494 |
|
| 495 |
+
# STEP 3: Retrieve relevant chunks using the question
|
| 496 |
+
retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 5})
|
| 497 |
+
relevant_docs = retriever.get_relevant_documents(question)
|
| 498 |
+
|
| 499 |
+
# STEP 4: Format the context
|
| 500 |
+
context = "\n\n".join([doc.page_content for doc in relevant_docs])
|
| 501 |
+
|
| 502 |
+
# STEP 5: Prompting
|
| 503 |
+
prompt_template = """
|
| 504 |
+
You are a helpful assistant that answers questions based on the context below.
|
| 505 |
+
|
| 506 |
+
Context:
|
| 507 |
+
{context}
|
| 508 |
+
|
| 509 |
+
Question: {question}
|
| 510 |
+
|
| 511 |
+
Answer:
|
| 512 |
+
""".strip()
|
| 513 |
+
|
| 514 |
+
prompt = prompt_template.format(context=context, question=question)
|
| 515 |
+
|
| 516 |
+
# STEP 6: Load the model and generate response
|
| 517 |
+
llm = HuggingFacePipeline.from_model_id(
|
| 518 |
+
model_id="MBZUAI/LaMini-T5-738M",
|
| 519 |
+
task="text2text-generation",
|
| 520 |
+
model_kwargs={"temperature": 0.3, "max_length": 256},
|
| 521 |
+
)
|
| 522 |
+
|
| 523 |
+
return llm.invoke(prompt)
|
| 524 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 525 |
|
| 526 |
# --- UI Layout ---
|
| 527 |
with st.sidebar:
|