Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
from fastapi import FastAPI, HTTPException
|
2 |
import os
|
3 |
-
from typing import List, Dict
|
4 |
from dotenv import load_dotenv
|
5 |
import logging
|
6 |
from pathlib import Path
|
@@ -21,9 +21,12 @@ from langgraph.graph import END
|
|
21 |
from langgraph.prebuilt import tools_condition
|
22 |
from langgraph.checkpoint.memory import MemorySaver
|
23 |
|
24 |
-
logging
|
|
|
|
|
25 |
logger = logging.getLogger(__name__)
|
26 |
|
|
|
27 |
load_dotenv()
|
28 |
GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
|
29 |
GROQ_API_KEY = os.getenv('GROQ_API_KEY')
|
@@ -41,36 +44,73 @@ class QASystem:
|
|
41 |
self.embeddings = None
|
42 |
self.client = None
|
43 |
self.pdf_dir = "pdfss"
|
|
|
44 |
|
45 |
def load_pdf_documents(self):
|
|
|
46 |
documents = []
|
47 |
pdf_dir = Path(self.pdf_dir)
|
48 |
|
49 |
if not pdf_dir.exists():
|
50 |
raise FileNotFoundError(f"PDF directory not found: {self.pdf_dir}")
|
51 |
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
try:
|
|
|
54 |
loader = PyPDFLoader(str(pdf_path))
|
55 |
-
|
56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
except Exception as e:
|
58 |
logger.error(f"Error loading PDF {pdf_path}: {str(e)}")
|
59 |
|
|
|
|
|
|
|
|
|
|
|
60 |
text_splitter = RecursiveCharacterTextSplitter(
|
61 |
chunk_size=1000,
|
62 |
-
chunk_overlap=
|
63 |
)
|
64 |
split_docs = text_splitter.split_documents(documents)
|
65 |
-
logger.info(f"Split documents into {len(split_docs)} chunks")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
return split_docs
|
67 |
|
68 |
def initialize_system(self):
|
|
|
69 |
try:
|
|
|
|
|
|
|
70 |
self.client = QdrantClient(":memory:")
|
|
|
71 |
|
|
|
72 |
try:
|
73 |
-
self.client.get_collection("pdf_data")
|
|
|
74 |
except Exception:
|
75 |
self.client.create_collection(
|
76 |
collection_name="pdf_data",
|
@@ -78,22 +118,32 @@ class QASystem:
|
|
78 |
)
|
79 |
logger.info("Created new collection: pdf_data")
|
80 |
|
|
|
81 |
self.embeddings = GoogleGenerativeAIEmbeddings(
|
82 |
model="models/embedding-001",
|
83 |
google_api_key=GOOGLE_API_KEY
|
84 |
)
|
|
|
85 |
|
|
|
86 |
self.vector_store = QdrantVectorStore(
|
87 |
client=self.client,
|
88 |
collection_name="pdf_data",
|
89 |
embeddings=self.embeddings,
|
90 |
)
|
|
|
91 |
|
|
|
92 |
documents = self.load_pdf_documents()
|
|
|
|
|
|
|
|
|
93 |
if documents:
|
94 |
try:
|
95 |
-
points = self.client.scroll(collection_name="pdf_data", limit=
|
96 |
if points:
|
|
|
97 |
self.client.delete(
|
98 |
collection_name="pdf_data",
|
99 |
points_selector=PointIdsList(
|
@@ -103,65 +153,104 @@ class QASystem:
|
|
103 |
except Exception as e:
|
104 |
logger.error(f"Error clearing vectors: {str(e)}")
|
105 |
|
|
|
|
|
106 |
self.vector_store.add_documents(documents)
|
107 |
-
logger.info(f"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
|
|
|
109 |
llm = ChatGroq(
|
110 |
model="llama3-8b-8192",
|
111 |
api_key=GROQ_API_KEY,
|
112 |
temperature=0.7
|
113 |
)
|
|
|
114 |
|
|
|
115 |
graph_builder = StateGraph(MessagesState)
|
|
|
116 |
|
117 |
-
# Define
|
|
|
|
|
118 |
def retrieve_docs(state: MessagesState):
|
|
|
119 |
# Get the most recent human message
|
120 |
human_messages = [m for m in state["messages"] if m.type == "human"]
|
121 |
if not human_messages:
|
|
|
122 |
return {"messages": state["messages"]}
|
123 |
|
124 |
user_query = human_messages[-1].content
|
125 |
-
logger.info(f"Retrieving documents for query: {user_query}")
|
|
|
|
|
|
|
|
|
|
|
126 |
|
127 |
# Query the vector store
|
128 |
try:
|
129 |
-
retrieved_docs =
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
|
131 |
-
# Create tool messages
|
132 |
tool_messages = []
|
133 |
for i, doc in enumerate(retrieved_docs):
|
|
|
|
|
|
|
134 |
tool_messages.append(
|
135 |
ToolMessage(
|
136 |
-
content=f"Document {i+1}: {doc.page_content}",
|
137 |
tool_call_id=f"retrieval_{i}"
|
138 |
)
|
139 |
)
|
140 |
|
141 |
-
logger.info(f"
|
142 |
return {"messages": state["messages"] + tool_messages}
|
143 |
|
144 |
except Exception as e:
|
145 |
logger.error(f"Error retrieving documents: {str(e)}")
|
146 |
return {"messages": state["messages"]}
|
147 |
|
148 |
-
#
|
149 |
def generate(state: MessagesState):
|
|
|
150 |
# Extract retrieved documents (tool messages)
|
151 |
tool_messages = [m for m in state["messages"] if m.type == "tool"]
|
152 |
|
153 |
# Collect context from retrieved documents
|
154 |
if tool_messages:
|
155 |
-
context = "\n".join([m.content for m in tool_messages])
|
156 |
logger.info(f"Using context from {len(tool_messages)} retrieved documents")
|
157 |
else:
|
158 |
-
context = "No specific mountain bicycle documentation available."
|
159 |
-
logger.
|
160 |
|
161 |
system_prompt = (
|
162 |
"You are an AI assistant embedded within the Interactive Electronic Technical Manual (IETM) for Mountain Cycles. "
|
163 |
-
"
|
164 |
-
"
|
|
|
|
|
|
|
165 |
f"\n\nContext from mountain bicycle documentation:\n{context}"
|
166 |
)
|
167 |
|
@@ -174,8 +263,14 @@ class QASystem:
|
|
174 |
logger.info(f"Sending query to LLM with {len(messages)} messages")
|
175 |
|
176 |
# Generate the response
|
177 |
-
|
178 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
179 |
|
180 |
# Add nodes to the graph
|
181 |
graph_builder.add_node("retrieve_docs", retrieve_docs)
|
@@ -186,22 +281,35 @@ class QASystem:
|
|
186 |
graph_builder.add_edge("retrieve_docs", "generate")
|
187 |
graph_builder.add_edge("generate", END)
|
188 |
|
|
|
189 |
self.memory = MemorySaver()
|
190 |
self.graph = graph_builder.compile(checkpointer=self.memory)
|
|
|
|
|
|
|
191 |
return True
|
192 |
|
193 |
except Exception as e:
|
194 |
logger.error(f"System initialization error: {str(e)}")
|
|
|
195 |
return False
|
196 |
|
197 |
-
def process_query(self, query: str) -> Dict[str,
|
198 |
"""Process a query and return a single final response"""
|
199 |
try:
|
200 |
-
|
201 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
202 |
thread_id = "abc123"
|
203 |
|
204 |
-
# Use invoke
|
205 |
final_state = self.graph.invoke(
|
206 |
{"messages": [HumanMessage(content=query)]},
|
207 |
config={"configurable": {"thread_id": thread_id}}
|
@@ -211,31 +319,35 @@ class QASystem:
|
|
211 |
ai_messages = [m for m in final_state["messages"] if m.type == "ai"]
|
212 |
|
213 |
if ai_messages:
|
|
|
214 |
# Return only the last AI message
|
215 |
return {
|
216 |
'content': ai_messages[-1].content,
|
217 |
'type': ai_messages[-1].type
|
218 |
}
|
|
|
|
|
219 |
return {
|
220 |
-
'content': "No response generated",
|
221 |
'type': 'error'
|
222 |
}
|
223 |
|
224 |
except Exception as e:
|
225 |
logger.error(f"Query processing error: {str(e)}")
|
226 |
return {
|
227 |
-
'content': f"
|
228 |
'type': 'error'
|
229 |
}
|
230 |
|
|
|
231 |
qa_system = QASystem()
|
232 |
-
|
233 |
-
logger.info("QA System Initialized Successfully")
|
234 |
-
else:
|
235 |
-
raise RuntimeError("Failed to initialize QA System")
|
236 |
|
237 |
@app.post("/query")
|
238 |
async def query_api(query: str):
|
239 |
"""API endpoint that returns a single response for a query"""
|
|
|
|
|
|
|
240 |
response = qa_system.process_query(query)
|
241 |
return {"response": response}
|
|
|
1 |
from fastapi import FastAPI, HTTPException
|
2 |
import os
|
3 |
+
from typing import List, Dict, Any
|
4 |
from dotenv import load_dotenv
|
5 |
import logging
|
6 |
from pathlib import Path
|
|
|
21 |
from langgraph.prebuilt import tools_condition
|
22 |
from langgraph.checkpoint.memory import MemorySaver
|
23 |
|
24 |
+
# Configure logging
|
25 |
+
logging.basicConfig(level=logging.INFO,
|
26 |
+
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
27 |
logger = logging.getLogger(__name__)
|
28 |
|
29 |
+
# Load environment variables
|
30 |
load_dotenv()
|
31 |
GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
|
32 |
GROQ_API_KEY = os.getenv('GROQ_API_KEY')
|
|
|
44 |
self.embeddings = None
|
45 |
self.client = None
|
46 |
self.pdf_dir = "pdfss"
|
47 |
+
self.is_initialized = False
|
48 |
|
49 |
def load_pdf_documents(self):
|
50 |
+
"""Load and process PDF documents from the pdf directory"""
|
51 |
documents = []
|
52 |
pdf_dir = Path(self.pdf_dir)
|
53 |
|
54 |
if not pdf_dir.exists():
|
55 |
raise FileNotFoundError(f"PDF directory not found: {self.pdf_dir}")
|
56 |
|
57 |
+
pdf_files = list(pdf_dir.glob("*.pdf"))
|
58 |
+
if not pdf_files:
|
59 |
+
logger.warning(f"No PDF files found in directory: {self.pdf_dir}")
|
60 |
+
return []
|
61 |
+
|
62 |
+
logger.info(f"Found {len(pdf_files)} PDF files to process")
|
63 |
+
|
64 |
+
for pdf_path in pdf_files:
|
65 |
try:
|
66 |
+
logger.info(f"Processing PDF: {pdf_path}")
|
67 |
loader = PyPDFLoader(str(pdf_path))
|
68 |
+
pdf_documents = loader.load()
|
69 |
+
|
70 |
+
# Add source information to metadata
|
71 |
+
for doc in pdf_documents:
|
72 |
+
if not hasattr(doc, 'metadata'):
|
73 |
+
doc.metadata = {}
|
74 |
+
doc.metadata['source'] = str(pdf_path.name)
|
75 |
+
|
76 |
+
documents.extend(pdf_documents)
|
77 |
+
logger.info(f"Loaded PDF: {pdf_path} - {len(pdf_documents)} pages/sections")
|
78 |
except Exception as e:
|
79 |
logger.error(f"Error loading PDF {pdf_path}: {str(e)}")
|
80 |
|
81 |
+
if not documents:
|
82 |
+
logger.warning("No documents were loaded from PDFs. Check the PDF directory and file formats.")
|
83 |
+
return []
|
84 |
+
|
85 |
+
# Split documents into smaller chunks for better retrieval
|
86 |
text_splitter = RecursiveCharacterTextSplitter(
|
87 |
chunk_size=1000,
|
88 |
+
chunk_overlap=200
|
89 |
)
|
90 |
split_docs = text_splitter.split_documents(documents)
|
91 |
+
logger.info(f"Split {len(documents)} documents into {len(split_docs)} chunks")
|
92 |
+
|
93 |
+
# Verify content of the first few chunks
|
94 |
+
for i, doc in enumerate(split_docs[:3]):
|
95 |
+
if i >= len(split_docs):
|
96 |
+
break
|
97 |
+
logger.info(f"Sample chunk {i+1} content preview: {doc.page_content[:100]}...")
|
98 |
+
|
99 |
return split_docs
|
100 |
|
101 |
def initialize_system(self):
|
102 |
+
"""Initialize the RAG system with vector store and LLM"""
|
103 |
try:
|
104 |
+
logger.info("Initializing QA System...")
|
105 |
+
|
106 |
+
# Initialize Qdrant client
|
107 |
self.client = QdrantClient(":memory:")
|
108 |
+
logger.info("Qdrant client initialized (in-memory)")
|
109 |
|
110 |
+
# Create or get collection
|
111 |
try:
|
112 |
+
collection_info = self.client.get_collection("pdf_data")
|
113 |
+
logger.info(f"Using existing collection: pdf_data")
|
114 |
except Exception:
|
115 |
self.client.create_collection(
|
116 |
collection_name="pdf_data",
|
|
|
118 |
)
|
119 |
logger.info("Created new collection: pdf_data")
|
120 |
|
121 |
+
# Initialize embeddings model
|
122 |
self.embeddings = GoogleGenerativeAIEmbeddings(
|
123 |
model="models/embedding-001",
|
124 |
google_api_key=GOOGLE_API_KEY
|
125 |
)
|
126 |
+
logger.info("Google AI Embeddings initialized")
|
127 |
|
128 |
+
# Initialize vector store
|
129 |
self.vector_store = QdrantVectorStore(
|
130 |
client=self.client,
|
131 |
collection_name="pdf_data",
|
132 |
embeddings=self.embeddings,
|
133 |
)
|
134 |
+
logger.info("Qdrant vector store initialized")
|
135 |
|
136 |
+
# Load documents
|
137 |
documents = self.load_pdf_documents()
|
138 |
+
if not documents:
|
139 |
+
logger.warning("No documents loaded. The system will continue but may not provide relevant responses.")
|
140 |
+
|
141 |
+
# Clear existing vectors if any
|
142 |
if documents:
|
143 |
try:
|
144 |
+
points = self.client.scroll(collection_name="pdf_data", limit=1000)[0]
|
145 |
if points:
|
146 |
+
logger.info(f"Clearing {len(points)} existing vectors from collection")
|
147 |
self.client.delete(
|
148 |
collection_name="pdf_data",
|
149 |
points_selector=PointIdsList(
|
|
|
153 |
except Exception as e:
|
154 |
logger.error(f"Error clearing vectors: {str(e)}")
|
155 |
|
156 |
+
# Add documents to vector store
|
157 |
+
logger.info(f"Adding {len(documents)} documents to vector store")
|
158 |
self.vector_store.add_documents(documents)
|
159 |
+
logger.info(f"Successfully added documents to vector store")
|
160 |
+
|
161 |
+
# Verify vector store has documents
|
162 |
+
try:
|
163 |
+
count = len(self.client.scroll(collection_name="pdf_data", limit=1)[0])
|
164 |
+
logger.info(f"Vector store contains points: {count > 0}")
|
165 |
+
except Exception as e:
|
166 |
+
logger.error(f"Error verifying vector store: {str(e)}")
|
167 |
|
168 |
+
# Initialize LLM
|
169 |
llm = ChatGroq(
|
170 |
model="llama3-8b-8192",
|
171 |
api_key=GROQ_API_KEY,
|
172 |
temperature=0.7
|
173 |
)
|
174 |
+
logger.info("Groq LLM initialized")
|
175 |
|
176 |
+
# Create LangGraph
|
177 |
graph_builder = StateGraph(MessagesState)
|
178 |
+
logger.info("Creating LangGraph for conversation flow")
|
179 |
|
180 |
+
# Define retrieval node (self reference for vector_store access)
|
181 |
+
vector_store_ref = self.vector_store
|
182 |
+
|
183 |
def retrieve_docs(state: MessagesState):
|
184 |
+
"""Node that retrieves relevant documents from the vector store"""
|
185 |
# Get the most recent human message
|
186 |
human_messages = [m for m in state["messages"] if m.type == "human"]
|
187 |
if not human_messages:
|
188 |
+
logger.warning("No human messages found in state")
|
189 |
return {"messages": state["messages"]}
|
190 |
|
191 |
user_query = human_messages[-1].content
|
192 |
+
logger.info(f"Retrieving documents for query: '{user_query}'")
|
193 |
+
|
194 |
+
# Check if vector store exists
|
195 |
+
if not vector_store_ref:
|
196 |
+
logger.error("Vector store not initialized or empty")
|
197 |
+
return {"messages": state["messages"]}
|
198 |
|
199 |
# Query the vector store
|
200 |
try:
|
201 |
+
retrieved_docs = vector_store_ref.similarity_search(user_query, k=3)
|
202 |
+
|
203 |
+
if not retrieved_docs:
|
204 |
+
logger.warning(f"No documents retrieved for query: '{user_query}'")
|
205 |
+
return {"messages": state["messages"]}
|
206 |
+
|
207 |
+
# Log what was actually retrieved
|
208 |
+
for i, doc in enumerate(retrieved_docs):
|
209 |
+
source = doc.metadata.get('source', 'Unknown') if hasattr(doc, 'metadata') else 'Unknown'
|
210 |
+
content_preview = doc.page_content[:100] + "..." if len(doc.page_content) > 100 else doc.page_content
|
211 |
+
logger.info(f"Retrieved doc {i+1} from {source}, preview: {content_preview}")
|
212 |
|
213 |
+
# Create tool messages with more detailed content
|
214 |
tool_messages = []
|
215 |
for i, doc in enumerate(retrieved_docs):
|
216 |
+
# Include source information if available
|
217 |
+
source_info = f" (Source: {doc.metadata.get('source', 'Unknown')})" if hasattr(doc, 'metadata') else ""
|
218 |
+
|
219 |
tool_messages.append(
|
220 |
ToolMessage(
|
221 |
+
content=f"Document {i+1}{source_info}: {doc.page_content}",
|
222 |
tool_call_id=f"retrieval_{i}"
|
223 |
)
|
224 |
)
|
225 |
|
226 |
+
logger.info(f"Created {len(tool_messages)} tool messages with retrieved content")
|
227 |
return {"messages": state["messages"] + tool_messages}
|
228 |
|
229 |
except Exception as e:
|
230 |
logger.error(f"Error retrieving documents: {str(e)}")
|
231 |
return {"messages": state["messages"]}
|
232 |
|
233 |
+
# Generate response using retrieved documents
|
234 |
def generate(state: MessagesState):
|
235 |
+
"""Node that generates a response using the LLM and retrieved documents"""
|
236 |
# Extract retrieved documents (tool messages)
|
237 |
tool_messages = [m for m in state["messages"] if m.type == "tool"]
|
238 |
|
239 |
# Collect context from retrieved documents
|
240 |
if tool_messages:
|
241 |
+
context = "\n\n".join([m.content for m in tool_messages])
|
242 |
logger.info(f"Using context from {len(tool_messages)} retrieved documents")
|
243 |
else:
|
244 |
+
context = "No specific mountain bicycle documentation available for this query."
|
245 |
+
logger.warning("No relevant documents retrieved, using default context")
|
246 |
|
247 |
system_prompt = (
|
248 |
"You are an AI assistant embedded within the Interactive Electronic Technical Manual (IETM) for Mountain Cycles. "
|
249 |
+
"Your primary role is to provide accurate technical information about mountain bicycles. "
|
250 |
+
"Always base your responses on the provided documentation. "
|
251 |
+
"If you don't find specific information in the provided context, clearly state that the information "
|
252 |
+
"is not available in the current documentation instead of making up details. "
|
253 |
+
"When responding, reference specific parts of the documentation."
|
254 |
f"\n\nContext from mountain bicycle documentation:\n{context}"
|
255 |
)
|
256 |
|
|
|
263 |
logger.info(f"Sending query to LLM with {len(messages)} messages")
|
264 |
|
265 |
# Generate the response
|
266 |
+
try:
|
267 |
+
response = llm.invoke(messages)
|
268 |
+
logger.info(f"LLM generated response successfully")
|
269 |
+
return {"messages": state["messages"] + [response]}
|
270 |
+
except Exception as e:
|
271 |
+
logger.error(f"Error generating response: {str(e)}")
|
272 |
+
error_message = SystemMessage(content=f"Error generating response: {str(e)}")
|
273 |
+
return {"messages": state["messages"] + [error_message]}
|
274 |
|
275 |
# Add nodes to the graph
|
276 |
graph_builder.add_node("retrieve_docs", retrieve_docs)
|
|
|
281 |
graph_builder.add_edge("retrieve_docs", "generate")
|
282 |
graph_builder.add_edge("generate", END)
|
283 |
|
284 |
+
# Initialize memory
|
285 |
self.memory = MemorySaver()
|
286 |
self.graph = graph_builder.compile(checkpointer=self.memory)
|
287 |
+
logger.info("Graph compiled successfully")
|
288 |
+
|
289 |
+
self.is_initialized = True
|
290 |
return True
|
291 |
|
292 |
except Exception as e:
|
293 |
logger.error(f"System initialization error: {str(e)}")
|
294 |
+
self.is_initialized = False
|
295 |
return False
|
296 |
|
297 |
+
def process_query(self, query: str) -> Dict[str, Any]:
|
298 |
"""Process a query and return a single final response"""
|
299 |
try:
|
300 |
+
if not self.is_initialized:
|
301 |
+
logger.error("System not initialized. Cannot process query.")
|
302 |
+
return {
|
303 |
+
'content': "Error: QA System not initialized properly",
|
304 |
+
'type': 'error'
|
305 |
+
}
|
306 |
+
|
307 |
+
logger.info(f"Processing query: '{query}'")
|
308 |
+
|
309 |
+
# Generate a thread ID (use a more sophisticated method for production)
|
310 |
thread_id = "abc123"
|
311 |
|
312 |
+
# Use invoke to get only the final result
|
313 |
final_state = self.graph.invoke(
|
314 |
{"messages": [HumanMessage(content=query)]},
|
315 |
config={"configurable": {"thread_id": thread_id}}
|
|
|
319 |
ai_messages = [m for m in final_state["messages"] if m.type == "ai"]
|
320 |
|
321 |
if ai_messages:
|
322 |
+
logger.info("Successfully generated response")
|
323 |
# Return only the last AI message
|
324 |
return {
|
325 |
'content': ai_messages[-1].content,
|
326 |
'type': ai_messages[-1].type
|
327 |
}
|
328 |
+
|
329 |
+
logger.warning("No AI message generated in response")
|
330 |
return {
|
331 |
+
'content': "No response could be generated for your query. Please try a different question.",
|
332 |
'type': 'error'
|
333 |
}
|
334 |
|
335 |
except Exception as e:
|
336 |
logger.error(f"Query processing error: {str(e)}")
|
337 |
return {
|
338 |
+
'content': f"Error processing your query: {str(e)}",
|
339 |
'type': 'error'
|
340 |
}
|
341 |
|
342 |
+
# Initialize the QA system
|
343 |
qa_system = QASystem()
|
344 |
+
initialization_success = qa_system.initialize_system()
|
|
|
|
|
|
|
345 |
|
346 |
@app.post("/query")
|
347 |
async def query_api(query: str):
|
348 |
"""API endpoint that returns a single response for a query"""
|
349 |
+
if not qa_system.is_initialized:
|
350 |
+
raise HTTPException(status_code=500, detail="QA System not initialized properly")
|
351 |
+
|
352 |
response = qa_system.process_query(query)
|
353 |
return {"response": response}
|