Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -429,7 +429,7 @@ from langchain_community.vectorstores import Chroma
|
|
| 429 |
from langchain_community.embeddings import SentenceTransformerEmbeddings
|
| 430 |
from langchain_community.llms import HuggingFacePipeline
|
| 431 |
from langchain.chains import RetrievalQA
|
| 432 |
-
from
|
| 433 |
from langchain.docstore.document import Document
|
| 434 |
|
| 435 |
# --- Streamlit Config ---
|
|
@@ -467,7 +467,6 @@ def create_vectorstore(text_chunks, embeddings):
|
|
| 467 |
shutil.rmtree(temp_dir)
|
| 468 |
os.makedirs(temp_dir, exist_ok=True)
|
| 469 |
|
| 470 |
-
# Wrap each chunk in a Document object
|
| 471 |
documents = [Document(page_content=chunk) for chunk in text_chunks]
|
| 472 |
db = Chroma.from_documents(documents, embedding=embeddings, persist_directory=temp_dir)
|
| 473 |
db.persist()
|
|
@@ -493,22 +492,29 @@ def process_question(question, full_text):
|
|
| 493 |
retriever = vectorstore.as_retriever()
|
| 494 |
|
| 495 |
llm = load_llm()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 496 |
qa = RetrievalQA.from_chain_type(
|
| 497 |
llm=llm,
|
| 498 |
retriever=retriever,
|
| 499 |
chain_type="stuff",
|
|
|
|
| 500 |
return_source_documents=False,
|
| 501 |
-
chain_type_kwargs={
|
| 502 |
-
"prompt": f"""You are a helpful assistant. Answer the user's question based only on the provided document content.
|
| 503 |
-
|
| 504 |
-
If the answer is clearly stated in the document, respond accurately and directly.
|
| 505 |
-
|
| 506 |
-
If not, say "The document does not provide enough information." Do not make things up.
|
| 507 |
-
|
| 508 |
-
Question: {question}
|
| 509 |
-
Context: {{context}}
|
| 510 |
-
Answer:"""
|
| 511 |
-
}
|
| 512 |
)
|
| 513 |
|
| 514 |
return qa.run(question)
|
|
@@ -551,4 +557,3 @@ if uploaded_file:
|
|
| 551 |
else:
|
| 552 |
st.info("Upload a PDF to begin.")
|
| 553 |
|
| 554 |
-
|
|
|
|
| 429 |
from langchain_community.embeddings import SentenceTransformerEmbeddings
|
| 430 |
from langchain_community.llms import HuggingFacePipeline
|
| 431 |
from langchain.chains import RetrievalQA
|
| 432 |
+
from langchain.prompts import PromptTemplate
|
| 433 |
from langchain.docstore.document import Document
|
| 434 |
|
| 435 |
# --- Streamlit Config ---
|
|
|
|
| 467 |
shutil.rmtree(temp_dir)
|
| 468 |
os.makedirs(temp_dir, exist_ok=True)
|
| 469 |
|
|
|
|
| 470 |
documents = [Document(page_content=chunk) for chunk in text_chunks]
|
| 471 |
db = Chroma.from_documents(documents, embedding=embeddings, persist_directory=temp_dir)
|
| 472 |
db.persist()
|
|
|
|
| 492 |
retriever = vectorstore.as_retriever()
|
| 493 |
|
| 494 |
llm = load_llm()
|
| 495 |
+
|
| 496 |
+
# ✅ Custom PromptTemplate
|
| 497 |
+
prompt_template = PromptTemplate(
|
| 498 |
+
input_variables=["context", "question"],
|
| 499 |
+
template="""
|
| 500 |
+
You are a helpful assistant. Answer the user's question based only on the provided document context below.
|
| 501 |
+
If the answer is in the context, answer it accurately. If not, say: "The document does not provide enough information."
|
| 502 |
+
|
| 503 |
+
Context:
|
| 504 |
+
{context}
|
| 505 |
+
|
| 506 |
+
Question:
|
| 507 |
+
{question}
|
| 508 |
+
|
| 509 |
+
Answer:"""
|
| 510 |
+
)
|
| 511 |
+
|
| 512 |
qa = RetrievalQA.from_chain_type(
|
| 513 |
llm=llm,
|
| 514 |
retriever=retriever,
|
| 515 |
chain_type="stuff",
|
| 516 |
+
chain_type_kwargs={"prompt": prompt_template},
|
| 517 |
return_source_documents=False,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 518 |
)
|
| 519 |
|
| 520 |
return qa.run(question)
|
|
|
|
| 557 |
else:
|
| 558 |
st.info("Upload a PDF to begin.")
|
| 559 |
|
|
|