Spaces:
Running
Running
Update app.py (#1)
Browse files- Update app.py (122a817fc3e71b709355262c7a64c37518c15c05)
app.py
CHANGED
|
@@ -4,6 +4,7 @@ from langchain_community.vectorstores import Chroma
|
|
| 4 |
from langchain_community.embeddings import HuggingFaceBgeEmbeddings
|
| 5 |
from groq import Groq
|
| 6 |
from dotenv import load_dotenv
|
|
|
|
| 7 |
|
| 8 |
# Initialize Streamlit page configuration
|
| 9 |
st.set_page_config(page_title="Medical Knowledge Assistant", layout="wide")
|
|
@@ -29,125 +30,117 @@ if not api_key:
|
|
| 29 |
# Initialize the app
|
| 30 |
st.title("Medical Knowledge Assistant")
|
| 31 |
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
model_kwargs = {'device': 'cpu'}
|
| 36 |
-
encode_kwargs = {'normalize_embeddings': False}
|
| 37 |
-
embeddings = HuggingFaceBgeEmbeddings(
|
| 38 |
-
model_name=model_name,
|
| 39 |
-
model_kwargs=model_kwargs,
|
| 40 |
-
encode_kwargs=encode_kwargs
|
| 41 |
-
)
|
| 42 |
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
st.
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
with st.sidebar:
|
| 53 |
-
st.write("API Key Loaded:", "Yes" if api_key else "No")
|
| 54 |
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
persist_directory=persist_directory,
|
| 65 |
-
embedding_function=embeddings
|
| 66 |
-
)
|
| 67 |
-
except Exception as e:
|
| 68 |
-
st.error(f"Error loading vector store: {e}")
|
| 69 |
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
# Initialize Groq client
|
| 78 |
-
client = Groq(api_key=api_key)
|
| 79 |
|
| 80 |
-
|
| 81 |
-
|
| 82 |
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
# Retrieve relevant documents
|
| 86 |
-
docs = retriever.get_relevant_documents(query)
|
| 87 |
-
context = "\n".join([doc.page_content for doc in docs])
|
| 88 |
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
"role": "system",
|
| 95 |
-
"content": (
|
| 96 |
-
"You are a knowledgeable medical assistant. For any medical term or disease, include comprehensive information covering: "
|
| 97 |
-
"definitions, types, historical background, major theories, known causes, and contributing risk factors. "
|
| 98 |
-
"Explain the genesis or theories on its origin, if applicable. Use a structured, thorough approach and keep language accessible. "
|
| 99 |
-
"provide symptoms, diagnosis, and treatment and post operative care , address all with indepth explanation , with specific details and step-by-step processes where relevant. "
|
| 100 |
-
"If the context does not adequately cover the user's question, respond with: 'I cannot provide an answer based on the available medical dataset.'"
|
| 101 |
-
)
|
| 102 |
-
},
|
| 103 |
-
{
|
| 104 |
-
"role": "system",
|
| 105 |
-
"content": (
|
| 106 |
-
"If the user asks for a medical explanation, ensure accuracy, don't include layman's terms if complex terms are used, "
|
| 107 |
-
"and organize responses in a structured way."
|
| 108 |
-
)
|
| 109 |
-
},
|
| 110 |
-
{
|
| 111 |
-
"role": "system",
|
| 112 |
-
"content": (
|
| 113 |
-
"When comparing two terms or conditions, provide a clear, concise, and structured comparison. Highlight key differences in their "
|
| 114 |
-
"definitions, symptoms, causes, diagnoses, and treatments with indepth explanation of each. If relevant, include any overlapping characteristics."
|
| 115 |
-
)
|
| 116 |
-
},
|
| 117 |
-
{
|
| 118 |
-
"role": "user",
|
| 119 |
-
"content": f"{context}\n\nQ: {query}\nA:"
|
| 120 |
-
}
|
| 121 |
-
],
|
| 122 |
-
temperature=0.7,
|
| 123 |
-
max_tokens=3000,
|
| 124 |
-
stream=True
|
| 125 |
-
)
|
| 126 |
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 142 |
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 151 |
|
| 152 |
-
|
| 153 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
from langchain_community.embeddings import HuggingFaceBgeEmbeddings
|
| 5 |
from groq import Groq
|
| 6 |
from dotenv import load_dotenv
|
| 7 |
+
import requests
|
| 8 |
|
| 9 |
# Initialize Streamlit page configuration
|
| 10 |
st.set_page_config(page_title="Medical Knowledge Assistant", layout="wide")
|
|
|
|
| 30 |
# Initialize the app
|
| 31 |
st.title("Medical Knowledge Assistant")
|
| 32 |
|
| 33 |
+
# Google Drive file ID (use your own file ID)
|
| 34 |
+
file_id = '1lVlF8dYsNFPzrNGqn7jiJos7qX49jmi0' # Replace with your Google Drive file ID
|
| 35 |
+
destination_path = '/tmp/Embedded_Med_books' # Temporary location to store the vector store
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
|
| 37 |
+
# Function to download file from Google Drive
|
| 38 |
+
def download_from_drive(file_id, destination_path):
|
| 39 |
+
"""Download the vector store file from Google Drive."""
|
| 40 |
+
url = f'https://drive.google.com/uc?export=download&id={file_id}'
|
| 41 |
+
response = requests.get(url)
|
| 42 |
+
if response.status_code == 200:
|
| 43 |
+
with open(destination_path, 'wb') as f:
|
| 44 |
+
f.write(response.content)
|
| 45 |
+
return destination_path
|
| 46 |
+
else:
|
| 47 |
+
st.error("Failed to download the file from Google Drive.")
|
| 48 |
+
return None
|
| 49 |
|
| 50 |
+
# Check if the vector store file exists, and download it if necessary
|
| 51 |
+
if not os.path.exists(destination_path):
|
| 52 |
+
st.warning("Downloading the vector store from Google Drive...")
|
| 53 |
+
download_from_drive(file_id, destination_path)
|
| 54 |
+
st.success("Vector store downloaded successfully!")
|
|
|
|
|
|
|
| 55 |
|
| 56 |
+
# Set up embeddings
|
| 57 |
+
model_name = "BAAI/bge-large-en"
|
| 58 |
+
model_kwargs = {'device': 'cpu'}
|
| 59 |
+
encode_kwargs = {'normalize_embeddings': False}
|
| 60 |
+
embeddings = HuggingFaceBgeEmbeddings(
|
| 61 |
+
model_name=model_name,
|
| 62 |
+
model_kwargs=model_kwargs,
|
| 63 |
+
encode_kwargs=encode_kwargs
|
| 64 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
|
| 66 |
+
# Load the vector store from the downloaded file
|
| 67 |
+
vector_store = Chroma(
|
| 68 |
+
persist_directory=destination_path,
|
| 69 |
+
embedding_function=embeddings
|
| 70 |
+
)
|
| 71 |
+
retriever = vector_store.as_retriever(search_kwargs={'k': 1})
|
|
|
|
|
|
|
|
|
|
| 72 |
|
| 73 |
+
# Initialize Groq client
|
| 74 |
+
client = Groq(api_key=api_key)
|
| 75 |
|
| 76 |
+
# Streamlit input
|
| 77 |
+
query = st.text_input("Enter your medical question here:")
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
+
def query_with_groq(query, retriever):
|
| 80 |
+
try:
|
| 81 |
+
# Retrieve relevant documents
|
| 82 |
+
docs = retriever.get_relevant_documents(query)
|
| 83 |
+
context = "\n".join([doc.page_content for doc in docs])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
|
| 85 |
+
# Call the Groq API with the query and context
|
| 86 |
+
completion = client.chat.completions.create(
|
| 87 |
+
model="llama3-70b-8192",
|
| 88 |
+
messages=[
|
| 89 |
+
{
|
| 90 |
+
"role": "system",
|
| 91 |
+
"content": (
|
| 92 |
+
"You are a knowledgeable medical assistant. For any medical term or disease, include comprehensive information covering: "
|
| 93 |
+
"definitions, types, historical background, major theories, known causes, and contributing risk factors. "
|
| 94 |
+
"Explain the genesis or theories on its origin, if applicable. Use a structured, thorough approach and keep language accessible. "
|
| 95 |
+
"provide symptoms, diagnosis, and treatment and post operative care , address all with indepth explanation , with specific details and step-by-step processes where relevant. "
|
| 96 |
+
"If the context does not adequately cover the user's question, respond with: 'I cannot provide an answer based on the available medical dataset.'"
|
| 97 |
+
)
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"role": "system",
|
| 101 |
+
"content": (
|
| 102 |
+
"If the user asks for a medical explanation, ensure accuracy, don't include layman's terms if complex terms are used, "
|
| 103 |
+
"and organize responses in a structured way."
|
| 104 |
+
)
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"role": "system",
|
| 108 |
+
"content": (
|
| 109 |
+
"When comparing two terms or conditions, provide a clear, concise, and structured comparison. Highlight key differences in their "
|
| 110 |
+
"definitions, symptoms, causes, diagnoses, and treatments with indepth explanation of each. If relevant, include any overlapping characteristics."
|
| 111 |
+
)
|
| 112 |
+
},
|
| 113 |
+
{
|
| 114 |
+
"role": "user",
|
| 115 |
+
"content": f"{context}\n\nQ: {query}\nA:"
|
| 116 |
+
}
|
| 117 |
+
],
|
| 118 |
+
temperature=0.7,
|
| 119 |
+
max_tokens=3000,
|
| 120 |
+
stream=True
|
| 121 |
+
)
|
| 122 |
|
| 123 |
+
# Create a placeholder for streaming response
|
| 124 |
+
response_container = st.empty()
|
| 125 |
+
response = ""
|
| 126 |
+
|
| 127 |
+
# Stream the response
|
| 128 |
+
for chunk in completion:
|
| 129 |
+
if chunk.choices[0].delta.content:
|
| 130 |
+
response += chunk.choices[0].delta.content
|
| 131 |
+
response_container.markdown(response)
|
| 132 |
+
|
| 133 |
+
return response
|
| 134 |
+
|
| 135 |
+
except Exception as e:
|
| 136 |
+
st.error(f"Error during query processing: {str(e)}")
|
| 137 |
+
return None
|
| 138 |
|
| 139 |
+
if st.button("Get Answer"):
|
| 140 |
+
if query:
|
| 141 |
+
with st.spinner("Processing your query..."):
|
| 142 |
+
answer = query_with_groq(query, retriever)
|
| 143 |
+
if answer:
|
| 144 |
+
st.success("Query processed successfully!")
|
| 145 |
+
else:
|
| 146 |
+
st.warning("Please enter a query.")
|