Update app.py

app.py

@@ -1,4 +1,6 @@
 import random
+import datetime
+import sys
 import os
 import torch
 import logging
@@ -19,7 +21,6 @@ current_dir = os.path.dirname(os.path.abspath(__file__))
 os.environ["MKL_THREADING_LAYER"] = "GNU"
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
-# Configuration
 CONFIG = {
     "model_name": "mims-harvard/TxAgent-T1-Llama-3.1-8B",
     "rag_model_name": "mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B",
@@ -33,94 +34,98 @@ CONFIG = {
     }
 }
 
-DESCRIPTION = '''
-<div>
-<h1 style="text-align: center;">TxAgent: An AI Agent for Therapeutic Reasoning Across a Universe of Tools</h1>
-</div>
-'''
-
-INTRO = """
-Precision therapeutics require multimodal adaptive models that provide personalized treatment recommendations. 
-We introduce TxAgent, an AI agent that leverages multi-step reasoning and real-time biomedical knowledge 
-retrieval across a toolbox of 211 expert-curated tools to navigate complex drug interactions, 
-contraindications, and patient-specific treatment strategies, delivering evidence-grounded therapeutic decisions.
+chat_css = """
+.gr-button { font-size: 20px !important; }
+.gr-button svg { width: 32px !important; height: 32px !important; }
 """
 
-LICENSE = """
-We welcome your feedback and suggestions to enhance your experience with TxAgent, and if you're interested 
-in collaboration, please email Marinka Zitnik and Shanghua Gao.
-
-### Medical Advice Disclaimer
-DISCLAIMER: THIS WEBSITE DOES NOT PROVIDE MEDICAL ADVICE
-The information, including but not limited to, text, graphics, images and other material contained on this 
-website are for informational purposes only. No material on this site is intended to be a substitute for 
-professional medical advice, diagnosis or treatment.
-"""
-
-PLACEHOLDER = """
-<div style="padding: 30px; text-align: center; display: flex; flex-direction: column; align-items: center;">
-   <h1 style="font-size: 28px; margin-bottom: 2px; opacity: 0.55;">TxAgent</h1>
-   <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.65;">Tips before using TxAgent:</p>
-   <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.55;">Please click clear🗑️ (top-right) to remove previous context before submitting a new question.</p>
-   <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.55;">Click retry🔄 (below message) to get multiple versions of the answer.</p>
-</div>
-"""
-
-def safe_load_embeddings(filepath: str):
-    """Handle embedding loading with fallbacks"""
+def safe_load_embeddings(filepath: str) -> any:
     try:
         return torch.load(filepath, weights_only=True)
     except Exception as e:
-        logger.warning(f"Secure load failed, trying without weights_only: {str(e)}")
+        logger.warning(f"Secure load failed, trying with weights_only=False: {str(e)}")
         try:
             return torch.load(filepath, weights_only=False)
         except Exception as e:
             logger.error(f"Failed to load embeddings: {str(e)}")
             return None
 
-def get_tools_from_universe(tooluniverse):
-    """Flexible tool extraction from ToolUniverse"""
-    if hasattr(tooluniverse, 'get_all_tools'):
-        return tooluniverse.get_all_tools()
-    elif hasattr(tooluniverse, 'tools'):
-        return tooluniverse.tools
-    elif hasattr(tooluniverse, 'list_tools'):
-        return tooluniverse.list_tools()
-    else:
-        logger.error("Could not find any tool access method in ToolUniverse")
-        # Try to load from files directly as fallback
-        tools = []
-        for tool_file in CONFIG["tool_files"].values():
-            if os.path.exists(tool_file):
-                with open(tool_file, 'r') as f:
-                    tools.extend(json.load(f))
-        return tools if tools else None
+def patch_embedding_loading():
+    try:
+        from txagent.toolrag import ToolRAGModel
+
+        def patched_load(self, tooluniverse):
+            try:
+                if not os.path.exists(CONFIG["embedding_filename"]):
+                    logger.error(f"Embedding file not found: {CONFIG['embedding_filename']}")
+                    return False
+
+                self.tool_desc_embedding = safe_load_embeddings(CONFIG["embedding_filename"])
+
+                if hasattr(tooluniverse, 'get_all_tools'):
+                    tools = tooluniverse.get_all_tools()
+                elif hasattr(tooluniverse, 'tools'):
+                    tools = tooluniverse.tools
+                else:
+                    logger.error("No method found to access tools from ToolUniverse")
+                    return False
+
+                current_count = len(tools)
+                embedding_count = len(self.tool_desc_embedding)
+
+                if current_count != embedding_count:
+                    logger.warning(f"Tool count mismatch (tools: {current_count}, embeddings: {embedding_count})")
+
+                    if current_count < embedding_count:
+                        self.tool_desc_embedding = self.tool_desc_embedding[:current_count]
+                        logger.info(f"Truncated embeddings to match {current_count} tools")
+                    else:
+                        last_embedding = self.tool_desc_embedding[-1]
+                        padding = [last_embedding] * (current_count - embedding_count)
+                        self.tool_desc_embedding = torch.cat([self.tool_desc_embedding] + padding)
+                        logger.info(f"Padded embeddings to match {current_count} tools")
+
+                return True
+
+            except Exception as e:
+                logger.error(f"Failed to load embeddings: {str(e)}")
+                return False
+
+        ToolRAGModel.load_tool_desc_embedding = patched_load
+        logger.info("Successfully patched embedding loading")
+
+    except Exception as e:
+        logger.error(f"Failed to patch embedding loading: {str(e)}")
+        raise
 
 def prepare_tool_files():
-    """Ensure tool files exist and are populated"""
     os.makedirs(os.path.join(current_dir, 'data'), exist_ok=True)
     if not os.path.exists(CONFIG["tool_files"]["new_tool"]):
-        logger.info("Generating tool list...")
+        logger.info("Generating tool list using ToolUniverse...")
         try:
             tu = ToolUniverse()
-            tools = get_tools_from_universe(tu)
-            if tools:
-                with open(CONFIG["tool_files"]["new_tool"], "w") as f:
-                    json.dump(tools, f, indent=2)
-                logger.info(f"Saved {len(tools)} tools")
+            if hasattr(tu, 'get_all_tools'):
+                tools = tu.get_all_tools()
+            elif hasattr(tu, 'tools'):
+                tools = tu.tools
             else:
-                logger.error("No tools could be loaded")
+                tools = []
+                logger.error("Could not access tools from ToolUniverse")
+
+            with open(CONFIG["tool_files"]["new_tool"], "w") as f:
+                json.dump(tools, f, indent=2)
+            logger.info(f"Saved {len(tools)} tools to {CONFIG['tool_files']['new_tool']}")
         except Exception as e:
-            logger.error(f"Tool file preparation failed: {str(e)}")
+            logger.error(f"Failed to prepare tool files: {str(e)}")
 
 def create_agent():
-    """Create and initialize the TxAgent with robust error handling"""
+    patch_embedding_loading()
     prepare_tool_files()
-    
+
     try:
         agent = TxAgent(
-            model_name=CONFIG["model_name"],
-            rag_model_name=CONFIG["rag_model_name"],
+            CONFIG["model_name"],
+            CONFIG["rag_model_name"],
             tool_files_dict=CONFIG["tool_files"],
             force_finish=True,
             enable_checker=True,
@@ -131,100 +136,71 @@ def create_agent():
         agent.init_model()
         return agent
     except Exception as e:
-        logger.error(f"Agent creation failed: {str(e)}")
+        logger.error(f"Failed to create agent: {str(e)}")
         raise
 
-def format_response(history, message):
-    """Properly format responses for Gradio Chatbot"""
-    if isinstance(message, (str, dict)):
-        return history + [[None, str(message)]]
-    elif hasattr(message, '__iter__'):
-        full_response = ""
-        for chunk in message:
-            if isinstance(chunk, dict):
-                full_response += chunk.get("content", "")
-            else:
-                full_response += str(chunk)
-        return history + [[None, full_response]]
-    return history + [[None, str(message)]]
+def respond(message, history, temperature, max_new_tokens, max_tokens, multi_agent, conversation, max_round):
+    updated_history = history + [{"role": "user", "content": message}]
+    print("\n==== DEBUG ====")
+    print("User Message:", message)
+    print("Full History:", updated_history)
+    print("================\n")
 
-def create_demo(agent):
-    """Create the Gradio interface with proper message handling"""
-    with gr.Blocks() as demo:
-        gr.Markdown(DESCRIPTION)
-        gr.Markdown(INTRO)
-        
-        chatbot = gr.Chatbot(
-            height=800,
-            label='TxAgent',
-            show_copy_button=True,
-            bubble_full_width=False
+    try:
+        # Ensure correct format for run_gradio_chat
+        formatted_history = [(m["role"], m["content"]) for m in updated_history]
+
+        response_generator = agent.run_gradio_chat(
+            formatted_history,
+            temperature,
+            max_new_tokens,
+            max_tokens,
+            multi_agent,
+            conversation,
+            max_round
         )
-        
-        msg = gr.Textbox(label="Input", placeholder="Type your question...")
-        clear = gr.ClearButton([msg, chatbot])
-        
-        def respond(message, chat_history):
-            try:
-                # Convert Gradio history to agent format
-                agent_history = []
-                for user_msg, bot_msg in chat_history:
-                    if user_msg:
-                        agent_history.append({"role": "user", "content": user_msg})
-                    if bot_msg:
-                        agent_history.append({"role": "assistant", "content": bot_msg})
-                
-                # Get response from agent
-                response = agent.run_gradio_chat(
-                    agent_history + [{"role": "user", "content": message}],
-                    temperature=0.3,
-                    max_new_tokens=1024,
-                    max_tokens=81920,
-                    multi_agent=False,
-                    conversation=[],
-                    max_round=30
-                )
-                
-                # Format the response properly
-                full_response = ""
-                for chunk in response:
-                    if isinstance(chunk, dict):
-                        full_response += chunk.get("content", "")
-                    else:
-                        full_response += str(chunk)
-                
-                return chat_history + [(message, full_response)]
-            
-            except Exception as e:
-                logger.error(f"Error in response handling: {str(e)}")
-                return chat_history + [(message, f"Error: {str(e)}")]
-
-        msg.submit(respond, [msg, chatbot], [chatbot])
-        clear.click(lambda: [], None, [chatbot])
-
-        # Add settings section
-        with gr.Accordion("Settings", open=False):
-            gr.Markdown("Adjust model parameters here")
-            
-            with gr.Row():
-                temperature = gr.Slider(0, 1, value=0.3, label="Temperature")
-                max_new_tokens = gr.Slider(128, 4096, value=1024, step=1, label="Max New Tokens")
-            
-            with gr.Row():
-                max_tokens = gr.Slider(128, 32000, value=81920, step=1, label="Max Tokens")
-                max_round = gr.Slider(1, 50, value=30, step=1, label="Max Round")
+    except Exception as e:
+        return history + [{"role": "user", "content": message}, {"role": "assistant", "content": f"Error: {str(e)}"}]
 
+    collected = ""
+    for chunk in response_generator:
+        if isinstance(chunk, dict):
+            collected += chunk.get("content", "")
+        else:
+            collected += str(chunk)
+
+    return history + [{"role": "user", "content": message}, {"role": "assistant", "content": collected}]
+
+def create_demo(agent):
+    with gr.Blocks(css=chat_css) as demo:
+        chatbot = gr.Chatbot(label="TxAgent", type="messages")
+        with gr.Row():
+            msg = gr.Textbox(label="Your question")
+        with gr.Row():
+            temp = gr.Slider(0, 1, value=0.3, label="Temperature")
+            max_new_tokens = gr.Slider(128, 4096, value=1024, label="Max New Tokens")
+            max_tokens = gr.Slider(128, 81920, value=81920, label="Max Total Tokens")
+            max_rounds = gr.Slider(1, 30, value=30, label="Max Rounds")
+            multi_agent = gr.Checkbox(label="Multi-Agent Mode")
+        with gr.Row():
+            submit = gr.Button("Ask TxAgent")
+
+        submit.click(
+            respond,
+            inputs=[msg, chatbot, temp, max_new_tokens, max_tokens, multi_agent, gr.State([]), max_rounds],
+            outputs=[chatbot]
+        )
     return demo
 
 def main():
-    """Main application entry point"""
     try:
+        global agent
         agent = create_agent()
         demo = create_demo(agent)
-        demo.launch(server_name="0.0.0.0", server_port=7860)
+        demo.launch()
     except Exception as e:
         logger.error(f"Application failed to start: {str(e)}")
         raise
 
 if __name__ == "__main__":
-    main()
+    main()