Spaces:
Runtime error
Runtime error
Update
#1
by afatuddin - opened
- Dockerfile +2 -3
- app.py → hf_app.py +57 -79
Dockerfile
CHANGED
|
@@ -9,8 +9,7 @@ RUN apt-get update && apt-get install -y \
|
|
| 9 |
&& rm -rf /var/lib/apt/lists/*
|
| 10 |
|
| 11 |
COPY requirements.txt ./
|
| 12 |
-
COPY
|
| 13 |
-
COPY .env ./
|
| 14 |
|
| 15 |
RUN pip3 install -r requirements.txt
|
| 16 |
|
|
@@ -18,4 +17,4 @@ EXPOSE 8501
|
|
| 18 |
|
| 19 |
HEALTHCHECK CMD curl --fail http://localhost:8501/_stcore/health
|
| 20 |
|
| 21 |
-
ENTRYPOINT ["streamlit", "run", "
|
|
|
|
| 9 |
&& rm -rf /var/lib/apt/lists/*
|
| 10 |
|
| 11 |
COPY requirements.txt ./
|
| 12 |
+
COPY src/ ./src/
|
|
|
|
| 13 |
|
| 14 |
RUN pip3 install -r requirements.txt
|
| 15 |
|
|
|
|
| 17 |
|
| 18 |
HEALTHCHECK CMD curl --fail http://localhost:8501/_stcore/health
|
| 19 |
|
| 20 |
+
ENTRYPOINT ["streamlit", "run", "src/streamlit_app.py", "--server.port=8501", "--server.address=0.0.0.0"]
|
app.py → hf_app.py
RENAMED
|
@@ -1,10 +1,6 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
from PyPDF2 import PdfReader
|
| 3 |
import os
|
| 4 |
-
from dotenv import load_dotenv
|
| 5 |
-
|
| 6 |
-
# Load environment variables from .env file
|
| 7 |
-
load_dotenv()
|
| 8 |
|
| 9 |
# --- UPDATED IMPORTS (LANGCHAIN v1.1 COMPATIBLE) ---
|
| 10 |
from langchain_core.documents import Document
|
|
@@ -15,13 +11,13 @@ from langchain_core.prompts import ChatPromptTemplate
|
|
| 15 |
from langchain_huggingface import HuggingFaceEndpoint, HuggingFaceEndpointEmbeddings, ChatHuggingFace
|
| 16 |
# ... rest of your code ...
|
| 17 |
@st.cache_data
|
| 18 |
-
def get_pdf_text(
|
| 19 |
"""
|
| 20 |
Extracts text from PDFs and creates Document objects with metadata.
|
| 21 |
Returns a list of Document objects, not a string.
|
| 22 |
"""
|
| 23 |
documents = []
|
| 24 |
-
for pdf in
|
| 25 |
pdf_reader = PdfReader(pdf)
|
| 26 |
pdf_filename = pdf.name
|
| 27 |
|
|
@@ -37,7 +33,7 @@ def get_pdf_text(_pdf_docs):
|
|
| 37 |
return documents
|
| 38 |
|
| 39 |
@st.cache_data
|
| 40 |
-
def get_text_chunks(
|
| 41 |
"""
|
| 42 |
Splits Document objects into chunks while preserving metadata.
|
| 43 |
"""
|
|
@@ -46,33 +42,29 @@ def get_text_chunks(_documents):
|
|
| 46 |
chunk_overlap=200
|
| 47 |
)
|
| 48 |
# split_documents ensures metadata (page num) is copied to new chunks
|
| 49 |
-
chunks = text_splitter.split_documents(
|
| 50 |
return chunks
|
| 51 |
|
| 52 |
# --- Hugging Face Model Functions ---
|
| 53 |
@st.cache_resource
|
| 54 |
def get_hf_embeddings():
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
st.error("HF_API_TOKEN not found in environment variables or secrets.toml.")
|
| 58 |
return None
|
| 59 |
return HuggingFaceEndpointEmbeddings(
|
| 60 |
model="BAAI/bge-small-en-v1.5",
|
| 61 |
-
huggingfacehub_api_token=
|
| 62 |
)
|
| 63 |
|
| 64 |
@st.cache_resource
|
| 65 |
def get_hf_llm():
|
| 66 |
-
|
| 67 |
-
if not hf_token:
|
| 68 |
return None
|
| 69 |
|
| 70 |
-
# Get model selection from session state
|
| 71 |
-
model_choice = st.session_state.get("selected_model", "mistralai/Mistral-7B-Instruct-v0.2")
|
| 72 |
-
|
| 73 |
llm = HuggingFaceEndpoint(
|
| 74 |
-
huggingfacehub_api_token=
|
| 75 |
-
repo_id=
|
|
|
|
| 76 |
temperature=0.3,
|
| 77 |
max_new_tokens=512,
|
| 78 |
task="conversational"
|
|
@@ -80,9 +72,9 @@ def get_hf_llm():
|
|
| 80 |
return ChatHuggingFace(llm=llm)
|
| 81 |
|
| 82 |
# --- Vector Store Functions (UPDATED) ---
|
| 83 |
-
def get_vector_store(
|
| 84 |
-
"""Creates FAISS vector store from Document objects
|
| 85 |
-
if not
|
| 86 |
st.error("No text chunks found.")
|
| 87 |
return
|
| 88 |
|
|
@@ -91,23 +83,25 @@ def get_vector_store(_text_chunks):
|
|
| 91 |
return
|
| 92 |
|
| 93 |
try:
|
| 94 |
-
#
|
| 95 |
-
vector_store = FAISS.from_documents(documents=
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
st.session_state.vector_store = vector_store
|
| 99 |
-
|
| 100 |
-
st.success("Vector store created and loaded successfully.")
|
| 101 |
except Exception as e:
|
| 102 |
st.error(f"Error creating vector store: {e}")
|
| 103 |
|
| 104 |
def load_faiss_index():
|
| 105 |
-
"""Load vector store from session state."""
|
| 106 |
try:
|
| 107 |
-
|
| 108 |
-
if
|
| 109 |
-
return
|
| 110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
except Exception as e:
|
| 112 |
# Silent fail on startup if index doesn't exist yet
|
| 113 |
return None
|
|
@@ -118,22 +112,19 @@ def get_conversational_chain():
|
|
| 118 |
if model is None:
|
| 119 |
return None
|
| 120 |
|
| 121 |
-
system_prompt = """
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
3. If the information needed to answer the question is not in the context, respond with: "I couldn't find this information in the provided documents. Please upload documents that contain the relevant information."
|
| 130 |
-
4. Be concise and direct in your answers.
|
| 131 |
-
5. If asking for clarification helps, you can ask follow-up questions.
|
| 132 |
-
6. Always cite which page or section your information comes from when possible."""
|
| 133 |
|
| 134 |
prompt = ChatPromptTemplate.from_messages([
|
| 135 |
("system", system_prompt),
|
| 136 |
-
|
|
|
|
| 137 |
])
|
| 138 |
|
| 139 |
# Create a simple chain using the pipe operator (|)
|
|
@@ -175,61 +166,48 @@ def user_input(user_question, db):
|
|
| 175 |
# Try to get string representation
|
| 176 |
st.write(str(response))
|
| 177 |
|
| 178 |
-
# 4. Display
|
|
|
|
|
|
|
| 179 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 180 |
|
| 181 |
except Exception as e:
|
| 182 |
st.error(f"Error: {e}")
|
| 183 |
-
import traceback
|
| 184 |
-
st.error(traceback.format_exc())
|
| 185 |
|
| 186 |
# --- Main App ---
|
| 187 |
def main():
|
| 188 |
st.set_page_config(page_title="Chat PDF (Hugging Face)")
|
| 189 |
-
st.header("Chat with PDF using
|
| 190 |
|
| 191 |
-
# Initialize session state variables
|
| 192 |
if "faiss_db" not in st.session_state:
|
| 193 |
st.session_state.faiss_db = None
|
| 194 |
-
|
| 195 |
-
|
| 196 |
-
if
|
| 197 |
-
|
| 198 |
|
| 199 |
with st.sidebar:
|
| 200 |
st.title("Menu:")
|
| 201 |
-
|
| 202 |
-
# Model Selection Dropdown
|
| 203 |
-
model_options = {
|
| 204 |
-
"Mistral 7B": "mistralai/Mistral-7B-Instruct-v0.2",
|
| 205 |
-
"Llama 3.2 3B": "meta-llama/Llama-3.2-3B-Instruct"
|
| 206 |
-
}
|
| 207 |
-
|
| 208 |
-
selected_model_name = st.selectbox(
|
| 209 |
-
"Choose Model:",
|
| 210 |
-
options=list(model_options.keys()),
|
| 211 |
-
index=0
|
| 212 |
-
)
|
| 213 |
-
|
| 214 |
-
st.session_state.selected_model = model_options[selected_model_name]
|
| 215 |
-
st.caption(f"Selected: `{st.session_state.selected_model}`")
|
| 216 |
-
|
| 217 |
-
st.divider()
|
| 218 |
-
|
| 219 |
pdf_docs = st.file_uploader("Upload PDFs", accept_multiple_files=True)
|
| 220 |
|
| 221 |
if st.button("Process PDFs"):
|
| 222 |
if pdf_docs:
|
| 223 |
with st.spinner("Processing..."):
|
| 224 |
-
# Clear previous cache to avoid using old vector stores
|
| 225 |
-
st.cache_data.clear()
|
| 226 |
-
st.cache_resource.clear()
|
| 227 |
-
|
| 228 |
# 1. Get Documents (not string)
|
| 229 |
raw_documents = get_pdf_text(pdf_docs)
|
| 230 |
# 2. Split Documents (preserves metadata)
|
| 231 |
text_chunks = get_text_chunks(raw_documents)
|
| 232 |
-
# 3. Create Vector Store
|
| 233 |
get_vector_store(text_chunks)
|
| 234 |
|
| 235 |
st.session_state.faiss_db = load_faiss_index()
|
|
|
|
| 1 |
import streamlit as st
|
| 2 |
from PyPDF2 import PdfReader
|
| 3 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
|
| 5 |
# --- UPDATED IMPORTS (LANGCHAIN v1.1 COMPATIBLE) ---
|
| 6 |
from langchain_core.documents import Document
|
|
|
|
| 11 |
from langchain_huggingface import HuggingFaceEndpoint, HuggingFaceEndpointEmbeddings, ChatHuggingFace
|
| 12 |
# ... rest of your code ...
|
| 13 |
@st.cache_data
|
| 14 |
+
def get_pdf_text(pdf_docs):
|
| 15 |
"""
|
| 16 |
Extracts text from PDFs and creates Document objects with metadata.
|
| 17 |
Returns a list of Document objects, not a string.
|
| 18 |
"""
|
| 19 |
documents = []
|
| 20 |
+
for pdf in pdf_docs:
|
| 21 |
pdf_reader = PdfReader(pdf)
|
| 22 |
pdf_filename = pdf.name
|
| 23 |
|
|
|
|
| 33 |
return documents
|
| 34 |
|
| 35 |
@st.cache_data
|
| 36 |
+
def get_text_chunks(documents):
|
| 37 |
"""
|
| 38 |
Splits Document objects into chunks while preserving metadata.
|
| 39 |
"""
|
|
|
|
| 42 |
chunk_overlap=200
|
| 43 |
)
|
| 44 |
# split_documents ensures metadata (page num) is copied to new chunks
|
| 45 |
+
chunks = text_splitter.split_documents(documents)
|
| 46 |
return chunks
|
| 47 |
|
| 48 |
# --- Hugging Face Model Functions ---
|
| 49 |
@st.cache_resource
|
| 50 |
def get_hf_embeddings():
|
| 51 |
+
if "HF_API_TOKEN" not in st.secrets:
|
| 52 |
+
st.error("HF_API_TOKEN not found in secrets.toml.")
|
|
|
|
| 53 |
return None
|
| 54 |
return HuggingFaceEndpointEmbeddings(
|
| 55 |
model="BAAI/bge-small-en-v1.5",
|
| 56 |
+
huggingfacehub_api_token=st.secrets["HF_API_TOKEN"]
|
| 57 |
)
|
| 58 |
|
| 59 |
@st.cache_resource
|
| 60 |
def get_hf_llm():
|
| 61 |
+
if "HF_API_TOKEN" not in st.secrets:
|
|
|
|
| 62 |
return None
|
| 63 |
|
|
|
|
|
|
|
|
|
|
| 64 |
llm = HuggingFaceEndpoint(
|
| 65 |
+
huggingfacehub_api_token=st.secrets["HF_API_TOKEN"],
|
| 66 |
+
repo_id="mistralai/Mistral-7B-Instruct-v0.2",
|
| 67 |
+
# repo_id="meta-llama/Llama-3.2-3B-Instruct",
|
| 68 |
temperature=0.3,
|
| 69 |
max_new_tokens=512,
|
| 70 |
task="conversational"
|
|
|
|
| 72 |
return ChatHuggingFace(llm=llm)
|
| 73 |
|
| 74 |
# --- Vector Store Functions (UPDATED) ---
|
| 75 |
+
def get_vector_store(text_chunks):
|
| 76 |
+
"""Creates FAISS vector store from Document objects."""
|
| 77 |
+
if not text_chunks:
|
| 78 |
st.error("No text chunks found.")
|
| 79 |
return
|
| 80 |
|
|
|
|
| 83 |
return
|
| 84 |
|
| 85 |
try:
|
| 86 |
+
# Use from_documents instead of from_texts to keep metadata
|
| 87 |
+
vector_store = FAISS.from_documents(documents=text_chunks, embedding=embeddings)
|
| 88 |
+
vector_store.save_local("faiss_index")
|
| 89 |
+
st.success("Vector store created and saved locally.")
|
|
|
|
|
|
|
|
|
|
| 90 |
except Exception as e:
|
| 91 |
st.error(f"Error creating vector store: {e}")
|
| 92 |
|
| 93 |
def load_faiss_index():
|
|
|
|
| 94 |
try:
|
| 95 |
+
embeddings = get_hf_embeddings()
|
| 96 |
+
if embeddings is None:
|
| 97 |
+
return None
|
| 98 |
+
|
| 99 |
+
db = FAISS.load_local(
|
| 100 |
+
"faiss_index",
|
| 101 |
+
embeddings,
|
| 102 |
+
allow_dangerous_deserialization=True
|
| 103 |
+
)
|
| 104 |
+
return db
|
| 105 |
except Exception as e:
|
| 106 |
# Silent fail on startup if index doesn't exist yet
|
| 107 |
return None
|
|
|
|
| 112 |
if model is None:
|
| 113 |
return None
|
| 114 |
|
| 115 |
+
system_prompt = """
|
| 116 |
+
You are a helpful AI assistant answering questions based on PDF documents.
|
| 117 |
+
If the answer is not in the context, say you don't know and do not give any text about the question, just say "I don't know."
|
| 118 |
+
Context:
|
| 119 |
+
{context}
|
| 120 |
+
**Note:** please do not answer questions that are not related to the context provided. If the context does not contain the answer, respond with "I don't know."
|
| 121 |
+
|
| 122 |
+
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
|
| 124 |
prompt = ChatPromptTemplate.from_messages([
|
| 125 |
("system", system_prompt),
|
| 126 |
+
# The user's question will be passed here as the "human" message
|
| 127 |
+
("human", "Question: {input}")
|
| 128 |
])
|
| 129 |
|
| 130 |
# Create a simple chain using the pipe operator (|)
|
|
|
|
| 166 |
# Try to get string representation
|
| 167 |
st.write(str(response))
|
| 168 |
|
| 169 |
+
# # 4. Display Page Numbers (New Feature)
|
| 170 |
+
# st.markdown("---")
|
| 171 |
+
# st.markdown("#### 📄 Source Pages:")
|
| 172 |
|
| 173 |
+
# # Use a set to avoid duplicate page numbers
|
| 174 |
+
# unique_sources = set()
|
| 175 |
+
# for doc in docs:
|
| 176 |
+
# page_num = doc.metadata.get("page", "Unknown")
|
| 177 |
+
# file_name = doc.metadata.get("source", "Unknown")
|
| 178 |
+
# unique_sources.add(f"{file_name} (Page {page_num})")
|
| 179 |
+
|
| 180 |
+
# # Display sources cleanly
|
| 181 |
+
# for source in unique_sources:
|
| 182 |
+
# st.markdown(f"- {source}")
|
| 183 |
|
| 184 |
except Exception as e:
|
| 185 |
st.error(f"Error: {e}")
|
|
|
|
|
|
|
| 186 |
|
| 187 |
# --- Main App ---
|
| 188 |
def main():
|
| 189 |
st.set_page_config(page_title="Chat PDF (Hugging Face)")
|
| 190 |
+
st.header("Chat with PDF using mistralai/Mistral-7B-Instruct-v0.2")
|
| 191 |
|
|
|
|
| 192 |
if "faiss_db" not in st.session_state:
|
| 193 |
st.session_state.faiss_db = None
|
| 194 |
+
|
| 195 |
+
# Attempt to load index on launch
|
| 196 |
+
if st.session_state.faiss_db is None and os.path.exists("faiss_index"):
|
| 197 |
+
st.session_state.faiss_db = load_faiss_index()
|
| 198 |
|
| 199 |
with st.sidebar:
|
| 200 |
st.title("Menu:")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 201 |
pdf_docs = st.file_uploader("Upload PDFs", accept_multiple_files=True)
|
| 202 |
|
| 203 |
if st.button("Process PDFs"):
|
| 204 |
if pdf_docs:
|
| 205 |
with st.spinner("Processing..."):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
# 1. Get Documents (not string)
|
| 207 |
raw_documents = get_pdf_text(pdf_docs)
|
| 208 |
# 2. Split Documents (preserves metadata)
|
| 209 |
text_chunks = get_text_chunks(raw_documents)
|
| 210 |
+
# 3. Create Vector Store
|
| 211 |
get_vector_store(text_chunks)
|
| 212 |
|
| 213 |
st.session_state.faiss_db = load_faiss_index()
|