Spaces:
Running
Running
Update rag_engine.py
Browse files- rag_engine.py +70 -52
rag_engine.py
CHANGED
@@ -11,6 +11,7 @@ import textwrap
|
|
11 |
import unicodedata
|
12 |
import streamlit as st
|
13 |
from utils import setup_gcp_auth, setup_openai_auth
|
|
|
14 |
|
15 |
# Force model to CPU for stability
|
16 |
os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
@@ -22,6 +23,7 @@ def initialize_session_state():
|
|
22 |
st.session_state.model = None
|
23 |
st.session_state.tokenizer = None
|
24 |
st.session_state.device = torch.device("cpu")
|
|
|
25 |
print("Initialized session state variables")
|
26 |
|
27 |
# Call the initialization function right away
|
@@ -38,7 +40,6 @@ def setup_gcp_client():
|
|
38 |
return bucket
|
39 |
except Exception as e:
|
40 |
print(f"β GCP client initialization error: {str(e)}")
|
41 |
-
st.error(f"GCP client initialization error: {str(e)}")
|
42 |
return None
|
43 |
|
44 |
# Setup OpenAI authentication
|
@@ -49,7 +50,6 @@ def setup_openai_client():
|
|
49 |
return True
|
50 |
except Exception as e:
|
51 |
print(f"β OpenAI client initialization error: {str(e)}")
|
52 |
-
st.error(f"OpenAI client initialization error: {str(e)}")
|
53 |
return False
|
54 |
|
55 |
# GCS Paths
|
@@ -66,32 +66,36 @@ local_metadata_file = "metadata.jsonl"
|
|
66 |
|
67 |
def load_model():
|
68 |
try:
|
69 |
-
#
|
70 |
-
if
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
#
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
|
|
|
|
|
|
|
|
93 |
|
94 |
-
return
|
95 |
except Exception as e:
|
96 |
print(f"β Error loading model: {str(e)}")
|
97 |
# Return None values instead of raising to avoid crashing
|
@@ -100,32 +104,41 @@ def load_model():
|
|
100 |
def download_file_from_gcs(bucket, gcs_path, local_path):
|
101 |
"""Download a file from GCS to local storage."""
|
102 |
try:
|
|
|
|
|
|
|
|
|
|
|
103 |
blob = bucket.blob(gcs_path)
|
104 |
blob.download_to_filename(local_path)
|
105 |
print(f"β
Downloaded {gcs_path} β {local_path}")
|
106 |
return True
|
107 |
except Exception as e:
|
108 |
print(f"β Error downloading {gcs_path}: {str(e)}")
|
109 |
-
st.error(f"Error downloading {gcs_path}: {str(e)}")
|
110 |
return False
|
111 |
|
112 |
def load_data_files():
|
|
|
|
|
|
|
|
|
|
|
113 |
# Initialize GCP and OpenAI clients
|
114 |
bucket = setup_gcp_client()
|
115 |
openai_initialized = setup_openai_client()
|
116 |
|
117 |
if not bucket or not openai_initialized:
|
118 |
-
|
119 |
return None, None, None
|
120 |
|
121 |
-
# Download necessary files
|
122 |
success = True
|
123 |
success &= download_file_from_gcs(bucket, faiss_index_file_gcs, local_faiss_index_file)
|
124 |
success &= download_file_from_gcs(bucket, text_chunks_file_gcs, local_text_chunks_file)
|
125 |
success &= download_file_from_gcs(bucket, metadata_file_gcs, local_metadata_file)
|
126 |
|
127 |
if not success:
|
128 |
-
|
129 |
return None, None, None
|
130 |
|
131 |
# Load FAISS index
|
@@ -133,7 +146,6 @@ def load_data_files():
|
|
133 |
faiss_index = faiss.read_index(local_faiss_index_file)
|
134 |
except Exception as e:
|
135 |
print(f"β Error loading FAISS index: {str(e)}")
|
136 |
-
st.error(f"Error loading FAISS index: {str(e)}")
|
137 |
return None, None, None
|
138 |
|
139 |
# Load text chunks
|
@@ -146,7 +158,6 @@ def load_data_files():
|
|
146 |
text_chunks[int(parts[0])] = (parts[1], parts[2], parts[3])
|
147 |
except Exception as e:
|
148 |
print(f"β Error loading text chunks: {str(e)}")
|
149 |
-
st.error(f"Error loading text chunks: {str(e)}")
|
150 |
return None, None, None
|
151 |
|
152 |
# Load metadata.jsonl for publisher information
|
@@ -158,10 +169,16 @@ def load_data_files():
|
|
158 |
metadata_dict[item["Title"]] = item # Store for easy lookup
|
159 |
except Exception as e:
|
160 |
print(f"β Error loading metadata: {str(e)}")
|
161 |
-
st.error(f"Error loading metadata: {str(e)}")
|
162 |
return None, None, None
|
163 |
|
164 |
print(f"β
FAISS index and text chunks loaded. {len(text_chunks)} passages available.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
165 |
return faiss_index, text_chunks, metadata_dict
|
166 |
|
167 |
def average_pool(last_hidden_states, attention_mask):
|
@@ -177,16 +194,17 @@ def get_embedding(text):
|
|
177 |
|
178 |
try:
|
179 |
# Ensure model initialization
|
180 |
-
if
|
181 |
tokenizer, model = load_model()
|
182 |
if model is None:
|
183 |
-
return np.zeros((1, 384), dtype=np.float32)
|
184 |
else:
|
185 |
tokenizer, model = st.session_state.tokenizer, st.session_state.model
|
186 |
|
|
|
187 |
input_text = f"query: {text}" if len(text) < 512 else f"passage: {text}"
|
188 |
|
189 |
-
# Explicitly specify truncation parameters
|
190 |
inputs = tokenizer(
|
191 |
input_text,
|
192 |
padding=True,
|
@@ -196,26 +214,25 @@ def get_embedding(text):
|
|
196 |
return_attention_mask=True
|
197 |
)
|
198 |
|
199 |
-
# Move to CPU explicitly
|
200 |
inputs = {k: v.to('cpu') for k, v in inputs.items()}
|
201 |
|
202 |
with torch.no_grad():
|
203 |
outputs = model(**inputs)
|
204 |
embeddings = average_pool(outputs.last_hidden_state, inputs['attention_mask'])
|
205 |
embeddings = nn.functional.normalize(embeddings, p=2, dim=1)
|
206 |
-
# Ensure we detach and move to numpy on CPU
|
207 |
embeddings = embeddings.detach().cpu().numpy()
|
208 |
|
209 |
# Explicitly clean up
|
210 |
-
del outputs
|
|
|
211 |
torch.cuda.empty_cache() if torch.cuda.is_available() else None
|
212 |
|
213 |
query_embedding_cache[text] = embeddings
|
214 |
return embeddings
|
215 |
except Exception as e:
|
216 |
print(f"β Embedding error: {str(e)}")
|
217 |
-
|
218 |
-
return np.zeros((1, 384), dtype=np.float32) # Changed from 1024 to 384 for e5-small-v2
|
219 |
|
220 |
def retrieve_passages(query, faiss_index, text_chunks, metadata_dict, top_k=5, similarity_threshold=0.5):
|
221 |
"""Retrieve top-k most relevant passages using FAISS with metadata."""
|
@@ -242,6 +259,7 @@ def retrieve_passages(query, faiss_index, text_chunks, metadata_dict, top_k=5, s
|
|
242 |
if clean_title in cited_titles:
|
243 |
continue
|
244 |
|
|
|
245 |
metadata_entry = metadata_dict.get(clean_title, {})
|
246 |
author = metadata_entry.get("Author", "Unknown")
|
247 |
publisher = metadata_entry.get("Publisher", "Unknown")
|
@@ -258,7 +276,6 @@ def retrieve_passages(query, faiss_index, text_chunks, metadata_dict, top_k=5, s
|
|
258 |
return retrieved_passages, retrieved_sources
|
259 |
except Exception as e:
|
260 |
print(f"β Error in retrieve_passages: {str(e)}")
|
261 |
-
st.error(f"Error in retrieve_passages: {str(e)}")
|
262 |
return [], []
|
263 |
|
264 |
def answer_with_llm(query, context=None, word_limit=100):
|
@@ -325,7 +342,6 @@ def answer_with_llm(query, context=None, word_limit=100):
|
|
325 |
|
326 |
except Exception as e:
|
327 |
print(f"β LLM API error: {str(e)}")
|
328 |
-
st.error(f"LLM API error: {str(e)}")
|
329 |
return "I apologize, but I'm unable to answer at the moment."
|
330 |
|
331 |
def format_citations(sources):
|
@@ -345,19 +361,21 @@ def process_query(query, top_k=5, word_limit=100):
|
|
345 |
print(f"\nπ Processing query: {query}")
|
346 |
|
347 |
# Load data files if not already loaded
|
348 |
-
|
349 |
-
st.session_state.faiss_index, st.session_state.text_chunks, st.session_state.metadata_dict = load_data_files()
|
350 |
-
st.session_state.data_loaded = True
|
351 |
|
352 |
# Check if data loaded successfully
|
353 |
-
if
|
354 |
-
return {
|
|
|
|
|
|
|
|
|
355 |
|
356 |
retrieved_context, retrieved_sources = retrieve_passages(
|
357 |
query,
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
top_k=top_k
|
362 |
)
|
363 |
|
|
|
11 |
import unicodedata
|
12 |
import streamlit as st
|
13 |
from utils import setup_gcp_auth, setup_openai_auth
|
14 |
+
import gc # Added for explicit garbage collection
|
15 |
|
16 |
# Force model to CPU for stability
|
17 |
os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
|
|
23 |
st.session_state.model = None
|
24 |
st.session_state.tokenizer = None
|
25 |
st.session_state.device = torch.device("cpu")
|
26 |
+
st.session_state.data_loaded = False
|
27 |
print("Initialized session state variables")
|
28 |
|
29 |
# Call the initialization function right away
|
|
|
40 |
return bucket
|
41 |
except Exception as e:
|
42 |
print(f"β GCP client initialization error: {str(e)}")
|
|
|
43 |
return None
|
44 |
|
45 |
# Setup OpenAI authentication
|
|
|
50 |
return True
|
51 |
except Exception as e:
|
52 |
print(f"β OpenAI client initialization error: {str(e)}")
|
|
|
53 |
return False
|
54 |
|
55 |
# GCS Paths
|
|
|
66 |
|
67 |
def load_model():
|
68 |
try:
|
69 |
+
# Check if model is already loaded
|
70 |
+
if st.session_state.model is not None and st.session_state.tokenizer is not None:
|
71 |
+
print("Model already loaded, reusing existing instance")
|
72 |
+
return st.session_state.tokenizer, st.session_state.model
|
73 |
+
|
74 |
+
# Force model to CPU - more stable than GPU for this use case
|
75 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
76 |
+
|
77 |
+
print("Loading tokenizer...")
|
78 |
+
tokenizer = AutoTokenizer.from_pretrained("intfloat/e5-small-v2")
|
79 |
+
|
80 |
+
print("Loading model...")
|
81 |
+
model = AutoModel.from_pretrained(
|
82 |
+
"intfloat/e5-small-v2",
|
83 |
+
torch_dtype=torch.float16 # Use half precision
|
84 |
+
)
|
85 |
+
|
86 |
+
# Move model to CPU explicitly
|
87 |
+
model = model.to('cpu')
|
88 |
+
model.eval()
|
89 |
+
torch.set_grad_enabled(False)
|
90 |
+
|
91 |
+
# Store in session state
|
92 |
+
st.session_state.tokenizer = tokenizer
|
93 |
+
st.session_state.model = model
|
94 |
+
st.session_state.model_initialized = True
|
95 |
+
|
96 |
+
print("β
Model loaded successfully")
|
97 |
|
98 |
+
return tokenizer, model
|
99 |
except Exception as e:
|
100 |
print(f"β Error loading model: {str(e)}")
|
101 |
# Return None values instead of raising to avoid crashing
|
|
|
104 |
def download_file_from_gcs(bucket, gcs_path, local_path):
|
105 |
"""Download a file from GCS to local storage."""
|
106 |
try:
|
107 |
+
# Check if file already exists locally
|
108 |
+
if os.path.exists(local_path):
|
109 |
+
print(f"File already exists locally: {local_path}")
|
110 |
+
return True
|
111 |
+
|
112 |
blob = bucket.blob(gcs_path)
|
113 |
blob.download_to_filename(local_path)
|
114 |
print(f"β
Downloaded {gcs_path} β {local_path}")
|
115 |
return True
|
116 |
except Exception as e:
|
117 |
print(f"β Error downloading {gcs_path}: {str(e)}")
|
|
|
118 |
return False
|
119 |
|
120 |
def load_data_files():
|
121 |
+
# Check if already loaded in session state
|
122 |
+
if hasattr(st.session_state, 'faiss_index') and st.session_state.faiss_index is not None:
|
123 |
+
print("Using cached data files from session state")
|
124 |
+
return st.session_state.faiss_index, st.session_state.text_chunks, st.session_state.metadata_dict
|
125 |
+
|
126 |
# Initialize GCP and OpenAI clients
|
127 |
bucket = setup_gcp_client()
|
128 |
openai_initialized = setup_openai_client()
|
129 |
|
130 |
if not bucket or not openai_initialized:
|
131 |
+
print("Failed to initialize required services")
|
132 |
return None, None, None
|
133 |
|
134 |
+
# Download necessary files
|
135 |
success = True
|
136 |
success &= download_file_from_gcs(bucket, faiss_index_file_gcs, local_faiss_index_file)
|
137 |
success &= download_file_from_gcs(bucket, text_chunks_file_gcs, local_text_chunks_file)
|
138 |
success &= download_file_from_gcs(bucket, metadata_file_gcs, local_metadata_file)
|
139 |
|
140 |
if not success:
|
141 |
+
print("Failed to download required files")
|
142 |
return None, None, None
|
143 |
|
144 |
# Load FAISS index
|
|
|
146 |
faiss_index = faiss.read_index(local_faiss_index_file)
|
147 |
except Exception as e:
|
148 |
print(f"β Error loading FAISS index: {str(e)}")
|
|
|
149 |
return None, None, None
|
150 |
|
151 |
# Load text chunks
|
|
|
158 |
text_chunks[int(parts[0])] = (parts[1], parts[2], parts[3])
|
159 |
except Exception as e:
|
160 |
print(f"β Error loading text chunks: {str(e)}")
|
|
|
161 |
return None, None, None
|
162 |
|
163 |
# Load metadata.jsonl for publisher information
|
|
|
169 |
metadata_dict[item["Title"]] = item # Store for easy lookup
|
170 |
except Exception as e:
|
171 |
print(f"β Error loading metadata: {str(e)}")
|
|
|
172 |
return None, None, None
|
173 |
|
174 |
print(f"β
FAISS index and text chunks loaded. {len(text_chunks)} passages available.")
|
175 |
+
|
176 |
+
# Store in session state
|
177 |
+
st.session_state.faiss_index = faiss_index
|
178 |
+
st.session_state.text_chunks = text_chunks
|
179 |
+
st.session_state.metadata_dict = metadata_dict
|
180 |
+
st.session_state.data_loaded = True
|
181 |
+
|
182 |
return faiss_index, text_chunks, metadata_dict
|
183 |
|
184 |
def average_pool(last_hidden_states, attention_mask):
|
|
|
194 |
|
195 |
try:
|
196 |
# Ensure model initialization
|
197 |
+
if not hasattr(st.session_state, 'model') or st.session_state.model is None:
|
198 |
tokenizer, model = load_model()
|
199 |
if model is None:
|
200 |
+
return np.zeros((1, 384), dtype=np.float32)
|
201 |
else:
|
202 |
tokenizer, model = st.session_state.tokenizer, st.session_state.model
|
203 |
|
204 |
+
# Prepare text
|
205 |
input_text = f"query: {text}" if len(text) < 512 else f"passage: {text}"
|
206 |
|
207 |
+
# Explicitly specify truncation parameters
|
208 |
inputs = tokenizer(
|
209 |
input_text,
|
210 |
padding=True,
|
|
|
214 |
return_attention_mask=True
|
215 |
)
|
216 |
|
217 |
+
# Move to CPU explicitly
|
218 |
inputs = {k: v.to('cpu') for k, v in inputs.items()}
|
219 |
|
220 |
with torch.no_grad():
|
221 |
outputs = model(**inputs)
|
222 |
embeddings = average_pool(outputs.last_hidden_state, inputs['attention_mask'])
|
223 |
embeddings = nn.functional.normalize(embeddings, p=2, dim=1)
|
|
|
224 |
embeddings = embeddings.detach().cpu().numpy()
|
225 |
|
226 |
# Explicitly clean up
|
227 |
+
del outputs, inputs
|
228 |
+
gc.collect()
|
229 |
torch.cuda.empty_cache() if torch.cuda.is_available() else None
|
230 |
|
231 |
query_embedding_cache[text] = embeddings
|
232 |
return embeddings
|
233 |
except Exception as e:
|
234 |
print(f"β Embedding error: {str(e)}")
|
235 |
+
return np.zeros((1, 384), dtype=np.float32)
|
|
|
236 |
|
237 |
def retrieve_passages(query, faiss_index, text_chunks, metadata_dict, top_k=5, similarity_threshold=0.5):
|
238 |
"""Retrieve top-k most relevant passages using FAISS with metadata."""
|
|
|
259 |
if clean_title in cited_titles:
|
260 |
continue
|
261 |
|
262 |
+
# Get metadata safely
|
263 |
metadata_entry = metadata_dict.get(clean_title, {})
|
264 |
author = metadata_entry.get("Author", "Unknown")
|
265 |
publisher = metadata_entry.get("Publisher", "Unknown")
|
|
|
276 |
return retrieved_passages, retrieved_sources
|
277 |
except Exception as e:
|
278 |
print(f"β Error in retrieve_passages: {str(e)}")
|
|
|
279 |
return [], []
|
280 |
|
281 |
def answer_with_llm(query, context=None, word_limit=100):
|
|
|
342 |
|
343 |
except Exception as e:
|
344 |
print(f"β LLM API error: {str(e)}")
|
|
|
345 |
return "I apologize, but I'm unable to answer at the moment."
|
346 |
|
347 |
def format_citations(sources):
|
|
|
361 |
print(f"\nπ Processing query: {query}")
|
362 |
|
363 |
# Load data files if not already loaded
|
364 |
+
faiss_index, text_chunks, metadata_dict = load_data_files()
|
|
|
|
|
365 |
|
366 |
# Check if data loaded successfully
|
367 |
+
if faiss_index is None or text_chunks is None or metadata_dict is None:
|
368 |
+
return {
|
369 |
+
"query": query,
|
370 |
+
"answer_with_rag": "β οΈ System error: Data files not loaded properly.",
|
371 |
+
"citations": "No citations available."
|
372 |
+
}
|
373 |
|
374 |
retrieved_context, retrieved_sources = retrieve_passages(
|
375 |
query,
|
376 |
+
faiss_index,
|
377 |
+
text_chunks,
|
378 |
+
metadata_dict,
|
379 |
top_k=top_k
|
380 |
)
|
381 |
|