Spaces:
Sleeping
Sleeping
Claude updates stuff
Browse files
app.py
CHANGED
@@ -63,22 +63,23 @@ def initialize_faiss_index(embeddings):
|
|
63 |
# Handle natural language queries
|
64 |
conversation_history = []
|
65 |
|
66 |
-
|
67 |
-
|
|
|
|
|
|
|
|
|
|
|
68 |
|
|
|
69 |
query_embedding = model.encode([query]).astype('float32')
|
70 |
-
|
71 |
-
# Search FAISS index
|
72 |
-
_, indices = faiss_index.search(query_embedding, 3) # Retrieve top 3 results
|
73 |
relevant_texts = [embeddings_texts[idx] for idx in indices[0]]
|
74 |
-
|
75 |
-
# Combine relevant texts and truncate if necessary
|
76 |
combined_text = "\n".join([text for text, _ in relevant_texts])
|
77 |
-
max_length = 500
|
78 |
if len(combined_text) > max_length:
|
79 |
combined_text = combined_text[:max_length] + "..."
|
80 |
|
81 |
-
# Generate a response using Gemini
|
82 |
try:
|
83 |
response = genai.generate_text(
|
84 |
model="models/text-bison-001",
|
@@ -87,71 +88,20 @@ def handle_query(query, faiss_index, embeddings_texts, model):
|
|
87 |
)
|
88 |
generated_text = response.result if response else "No response generated."
|
89 |
except Exception as e:
|
90 |
-
|
91 |
-
generated_text = "An error occurred while generating the response."
|
92 |
-
|
93 |
-
# Update conversation history
|
94 |
-
conversation_history.append(f"User: {query}")
|
95 |
-
conversation_history.append(f"System: {generated_text}")
|
96 |
|
97 |
-
# Extract sources
|
98 |
sources = [url for _, url in relevant_texts]
|
99 |
-
|
100 |
return generated_text, sources
|
101 |
|
102 |
-
def generate_concise_response(prompt, context):
|
103 |
-
try:
|
104 |
-
response = genai.generate_text(
|
105 |
-
model="models/text-bison-001",
|
106 |
-
prompt=f"{prompt}\n\nContext: {context}\n\nAnswer:",
|
107 |
-
max_output_tokens=200
|
108 |
-
)
|
109 |
-
return response.result if response else "No response generated."
|
110 |
-
except Exception as e:
|
111 |
-
print(f"Error generating concise response: {e}")
|
112 |
-
return "An error occurred while generating the concise response."
|
113 |
-
|
114 |
-
# Main function to execute the pipeline
|
115 |
def chatbot(message, history):
|
116 |
-
global conversation_history
|
117 |
-
|
118 |
-
lecture_notes = fetch_lecture_notes()
|
119 |
-
model_architectures = fetch_model_architectures()
|
120 |
-
|
121 |
-
all_texts = lecture_notes + [model_architectures]
|
122 |
-
|
123 |
-
# Load the SentenceTransformers model
|
124 |
-
embedding_model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
|
125 |
-
|
126 |
-
embeddings = create_embeddings(all_texts, embedding_model)
|
127 |
-
|
128 |
-
# Initialize FAISS index
|
129 |
-
faiss_index = initialize_faiss_index(np.array(embeddings))
|
130 |
-
|
131 |
response, sources = handle_query(message, faiss_index, all_texts, embedding_model)
|
132 |
-
|
133 |
-
print("Response:", response)
|
134 |
total_text = response if response else "No response generated."
|
135 |
if sources:
|
136 |
-
|
137 |
-
|
138 |
-
for source in sources:
|
139 |
-
relevant_source += source + "\n"
|
140 |
-
total_text += "\n\nSources:\n" + relevant_source
|
141 |
-
else:
|
142 |
-
print("Sources: None of the provided sources were used.")
|
143 |
-
print("----")
|
144 |
-
|
145 |
-
# Generate a concise and relevant summary using Gemini
|
146 |
-
prompt = "Summarize the user queries so far"
|
147 |
-
user_queries_summary = " ".join([message])
|
148 |
-
concise_response = generate_concise_response(prompt, user_queries_summary)
|
149 |
-
print("Concise Response:")
|
150 |
-
print(concise_response)
|
151 |
|
152 |
-
# Update conversation history with the new user message and system response
|
153 |
history.append((message, total_text))
|
154 |
-
|
155 |
return history
|
156 |
|
157 |
iface = gr.ChatInterface(
|
@@ -163,11 +113,11 @@ iface = gr.ChatInterface(
|
|
163 |
"Explain the transformer architecture.",
|
164 |
"Tell me about datasets used to train LLMs.",
|
165 |
"How are LLM training datasets cleaned and preprocessed?",
|
166 |
-
"Summarize the user queries so far"
|
167 |
],
|
168 |
retry_btn="Regenerate",
|
169 |
undo_btn="Undo",
|
170 |
clear_btn="Clear",
|
|
|
171 |
)
|
172 |
|
173 |
if __name__ == "__main__":
|
|
|
63 |
# Handle natural language queries
|
64 |
conversation_history = []
|
65 |
|
66 |
+
# Global variables
|
67 |
+
lecture_notes = fetch_lecture_notes()
|
68 |
+
model_architectures = fetch_model_architectures()
|
69 |
+
all_texts = lecture_notes + [model_architectures]
|
70 |
+
embedding_model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
|
71 |
+
embeddings = create_embeddings(all_texts, embedding_model)
|
72 |
+
faiss_index = initialize_faiss_index(np.array(embeddings))
|
73 |
|
74 |
+
def handle_query(query, faiss_index, embeddings_texts, model):
|
75 |
query_embedding = model.encode([query]).astype('float32')
|
76 |
+
_, indices = faiss_index.search(query_embedding, 3)
|
|
|
|
|
77 |
relevant_texts = [embeddings_texts[idx] for idx in indices[0]]
|
|
|
|
|
78 |
combined_text = "\n".join([text for text, _ in relevant_texts])
|
79 |
+
max_length = 500
|
80 |
if len(combined_text) > max_length:
|
81 |
combined_text = combined_text[:max_length] + "..."
|
82 |
|
|
|
83 |
try:
|
84 |
response = genai.generate_text(
|
85 |
model="models/text-bison-001",
|
|
|
88 |
)
|
89 |
generated_text = response.result if response else "No response generated."
|
90 |
except Exception as e:
|
91 |
+
generated_text = f"An error occurred while generating the response: {str(e)}"
|
|
|
|
|
|
|
|
|
|
|
92 |
|
|
|
93 |
sources = [url for _, url in relevant_texts]
|
|
|
94 |
return generated_text, sources
|
95 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
def chatbot(message, history):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
97 |
response, sources = handle_query(message, faiss_index, all_texts, embedding_model)
|
98 |
+
|
|
|
99 |
total_text = response if response else "No response generated."
|
100 |
if sources:
|
101 |
+
relevant_source = "\n".join(sources)
|
102 |
+
total_text += f"\n\nSources:\n{relevant_source}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
103 |
|
|
|
104 |
history.append((message, total_text))
|
|
|
105 |
return history
|
106 |
|
107 |
iface = gr.ChatInterface(
|
|
|
113 |
"Explain the transformer architecture.",
|
114 |
"Tell me about datasets used to train LLMs.",
|
115 |
"How are LLM training datasets cleaned and preprocessed?",
|
|
|
116 |
],
|
117 |
retry_btn="Regenerate",
|
118 |
undo_btn="Undo",
|
119 |
clear_btn="Clear",
|
120 |
+
cache_examples=False, # Disable example caching to avoid file-related errors
|
121 |
)
|
122 |
|
123 |
if __name__ == "__main__":
|