Update app.py

app.py

@@ -1,16 +1,21 @@
- import ThreadPoolExecutor, as_completed
+import sys
+import os
+import pandas as pd
+import pdfplumber
+import json
+import gradio as gr
+from typing import List
+from concurrent.futures import ThreadPoolExecutor, as_completed
 import hashlib
 import shutil
 import time
-from threading import Thread, Lock
 import re
-import tempfile
-import threading
+import psutil
+import subprocess
 
 # ---------------------------------------------------------------------------------------
-# Setup persistent directories for Hugging Face Spaces
+# Persistent directory for Hugging Face Spaces
 # ---------------------------------------------------------------------------------------
-# Use a persistent cache directory (adjust the path as needed based on your HF Space settings)
 persistent_dir = "/data/hf_cache"
 os.makedirs(persistent_dir, exist_ok=True)
 
@@ -23,25 +28,23 @@ vllm_cache_dir = os.path.join(persistent_dir, "vllm_cache")
 for directory in [model_cache_dir, tool_cache_dir, file_cache_dir, report_dir, vllm_cache_dir]:
     os.makedirs(directory, exist_ok=True)
 
-# Set environment variables so that model and transformers caches point to persistent storage.
 os.environ["HF_HOME"] = model_cache_dir
 os.environ["TRANSFORMERS_CACHE"] = model_cache_dir
 os.environ["VLLM_CACHE_DIR"] = vllm_cache_dir
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
 
-# Append the local source path if needed
+# ---------------------------------------------------------------------------------------
+# Add src to path
+# ---------------------------------------------------------------------------------------
 current_dir = os.path.dirname(os.path.abspath(__file__))
 src_path = os.path.abspath(os.path.join(current_dir, "src"))
 sys.path.insert(0, src_path)
 
-# ---------------------------------------------------------------------------------------
-# Import the TxAgent from your tool package
-# ---------------------------------------------------------------------------------------
 from txagent.txagent import TxAgent
 
 # ---------------------------------------------------------------------------------------
-# Define constants and helper functions
+# Helper functions
 # ---------------------------------------------------------------------------------------
 MEDICAL_KEYWORDS = {
     'diagnosis', 'assessment', 'plan', 'results', 'medications',
@@ -59,11 +62,9 @@ def extract_priority_pages(file_path: str, max_pages: int = 20) -> str:
     try:
         text_chunks = []
         with pdfplumber.open(file_path) as pdf:
-            # Process first three pages always
             for i, page in enumerate(pdf.pages[:3]):
                 text = page.extract_text() or ""
                 text_chunks.append(f"=== Page {i+1} ===\n{text.strip()}")
-            # Process subsequent pages only if they contain key medical keywords
             for i, page in enumerate(pdf.pages[3:max_pages], start=4):
                 page_text = page.extract_text() or ""
                 if any(re.search(rf'\b{kw}\b', page_text.lower()) for kw in MEDICAL_KEYWORDS):
@@ -83,7 +84,6 @@ def convert_file_to_json(file_path: str, file_type: str) -> str:
         if file_type == "pdf":
             text = extract_priority_pages(file_path)
             result = json.dumps({"filename": os.path.basename(file_path), "content": text, "status": "initial"})
-            Thread(target=full_pdf_processing, args=(file_path, h)).start()
         elif file_type == "csv":
             df = pd.read_csv(file_path, encoding_errors="replace", header=None, dtype=str, 
                              skip_blank_lines=False, on_bad_lines="skip")
@@ -104,34 +104,34 @@ def convert_file_to_json(file_path: str, file_type: str) -> str:
     except Exception as e:
         return json.dumps({"error": f"Error processing {os.path.basename(file_path)}: {str(e)}"})
 
-def full_pdf_processing(file_path: str, file_hash_value: str):
+def log_system_usage(tag=""):
     try:
-        cache_path = os.path.join(file_cache_dir, f"{file_hash_value}_full.json")
-        if os.path.exists(cache_path):
-            return
-        with pdfplumber.open(file_path) as pdf:
-            full_text = "\n".join([f"=== Page {i+1} ===\n{(page.extract_text() or '').strip()}" 
-                                   for i, page in enumerate(pdf.pages)])
-        result = json.dumps({"filename": os.path.basename(file_path), "content": full_text, "status": "complete"})
-        with open(cache_path, "w", encoding="utf-8") as f:
-            f.write(result)
-        with open(os.path.join(report_dir, f"{file_hash_value}_report.txt"), "w", encoding="utf-8") as out:
-            out.write(full_text)
+        cpu_percent = psutil.cpu_percent(interval=1)
+        mem = psutil.virtual_memory()
+        print(f"[{tag}] 🧠 CPU: {cpu_percent}% | RAM: {mem.used // (1024**2)}MB / {mem.total // (1024**2)}MB")
+        result = subprocess.run(
+            ["nvidia-smi", "--query-gpu=memory.used,memory.total,utilization.gpu", "--format=csv,nounits,noheader"],
+            capture_output=True,
+            text=True,
+        )
+        if result.returncode == 0:
+            mem_used, mem_total, util = result.stdout.strip().split(", ")
+            print(f"[{tag}] ⚡ GPU: {mem_used}MB / {mem_total}MB | Utilization: {util}%")
+        else:
+            print(f"[{tag}] ⚡ GPU info not available.")
     except Exception as e:
-        print(f"Background processing failed: {str(e)}")
-
-# ---------------------------------------------------------------------------------------
-# Global agent variable and thread-safe lock for background model loading
-# ---------------------------------------------------------------------------------------
-agent = None
-agent_lock = Lock()
+        print(f"[{tag}] ⚠️ Failed to log system usage: {e}")
 
 def init_agent():
+    print("🔁 Initializing TxAgent...")
+    log_system_usage("Before Model Load")
+
     default_tool_path = os.path.abspath("data/new_tool.json")
     target_tool_path = os.path.join(tool_cache_dir, "new_tool.json")
     if not os.path.exists(target_tool_path):
         shutil.copy(default_tool_path, target_tool_path)
-    new_agent = TxAgent(
+
+    agent = TxAgent(
         model_name="mims-harvard/TxAgent-T1-Llama-3.1-8B",
         rag_model_name="mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B",
         tool_files_dict={"new_tool": target_tool_path},
@@ -141,24 +141,21 @@ def init_agent():
         seed=100,
         additional_default_tools=[],
     )
-    new_agent.init_model()
-    return new_agent
+    agent.init_model()
+    log_system_usage("After Model Load")
 
-def load_agent_in_background():
-    global agent
-    with agent_lock:
-        if agent is None:
-            print("Initializing agent in background (this may take a while)...")
-            agent = init_agent()
-            print("Agent initialization complete.")
+    print("✅ TxAgent is ready.")
+    print("📦 Cached model files:")
+    for root, _, files in os.walk(model_cache_dir):
+        for file in files:
+            print(os.path.join(root, file))
 
-# Start background agent loading at startup
-threading.Thread(target=load_agent_in_background, daemon=True).start()
+    return agent
 
 # ---------------------------------------------------------------------------------------
-# Define the Gradio UI
+# Gradio UI
 # ---------------------------------------------------------------------------------------
-def create_ui():
+def create_ui(agent):
     with gr.Blocks(theme=gr.themes.Soft()) as demo:
         gr.Markdown("""
         <h1 style='text-align: center;'>🩺 Clinical Oversight Assistant</h1>
@@ -173,20 +170,10 @@ def create_ui():
         download_output = gr.File(label="Download Full Report")
 
         def analyze_potential_oversights(message: str, history: list, files: list):
-            global agent
-            # Append user and interim assistant message
-            history = history + [
-                {"role": "user", "content": message},
-                {"role": "assistant", "content": "⏳ Analyzing records for potential oversights..."}
-            ]
+            history = history + [{"role": "user", "content": message},
+                                 {"role": "assistant", "content": "⏳ Analyzing records for potential oversights..."}]
             yield history, None
 
-            if agent is None:
-                history.append({"role": "assistant", 
-                                "content": "🕒 The model is still loading. Please wait a moment and try again."})
-                yield history, None
-                return
-
             extracted_data = ""
             file_hash_value = ""
             if files and isinstance(files, list):
@@ -195,13 +182,10 @@ def create_ui():
                         executor.submit(convert_file_to_json, f.name, f.name.split(".")[-1].lower())
                         for f in files if hasattr(f, 'name')
                     ]
-                    results = []
-                    for future in as_completed(futures):
-                        results.append(sanitize_utf8(future.result()))
+                    results = [sanitize_utf8(f.result()) for f in as_completed(futures)]
                     extracted_data = "\n".join(results)
                     file_hash_value = file_hash(files[0].name) if hasattr(files[0], 'name') else ""
 
-            # Truncate extracted data to avoid token overflow
             max_extracted_chars = 12000
             truncated_data = extracted_data[:max_extracted_chars]
 
@@ -216,10 +200,8 @@ Medical Records:
 
 ### Potential Oversights:
 """
-
             response = ""
             try:
-                # Stream agent responses and update the last message in the conversation with each chunk.
                 for chunk in agent.run_gradio_chat(
                     message=analysis_prompt,
                     history=[],
@@ -229,18 +211,16 @@ Medical Records:
                     call_agent=False,
                     conversation=[]
                 ):
-                    if chunk is None:
-                        continue
+                    if chunk is None: continue
                     if isinstance(chunk, str):
                         response += chunk
                     elif isinstance(chunk, list):
                         response += "".join([c.content for c in chunk if hasattr(c, 'content')])
                     cleaned = response.replace("[TOOL_CALLS]", "").strip()
-                    # Update the assistant message (last item in history) with the latest accumulated answer
                     history[-1] = {"role": "assistant", "content": cleaned}
                     yield history, None
             except Exception as agent_error:
-                history[-1] = {"role": "assistant", "content": f"❌ Analysis failed during processing: {str(agent_error)}"}
+                history[-1] = {"role": "assistant", "content": f"❌ Analysis failed: {str(agent_error)}"}
                 yield history, None
                 return
 
@@ -248,7 +228,6 @@ Medical Records:
             if not final_output:
                 final_output = "No clear oversights identified. Recommend comprehensive review."
 
-            # Update the assistant's message with the final output
             history[-1] = {"role": "assistant", "content": final_output}
 
             report_path = None
@@ -265,15 +244,20 @@ Medical Records:
         msg_input.submit(analyze_potential_oversights,
                          inputs=[msg_input, gr.State([]), file_upload],
                          outputs=[chatbot, download_output])
-        gr.Examples([["What might have been missed in this patient's treatment?"],
-                     ["Are there any medication conflicts in these records?"],
-                     ["What abnormal results require follow-up?"]],
-                    inputs=msg_input)
+        gr.Examples([
+            ["What might have been missed in this patient's treatment?"],
+            ["Are there any medication conflicts in these records?"],
+            ["What abnormal results require follow-up?"]],
+            inputs=msg_input)
     return demo
 
+# ---------------------------------------------------------------------------------------
+# Launch
+# ---------------------------------------------------------------------------------------
 if __name__ == "__main__":
-    print("Launching interface...")
-    demo = create_ui()
+    print("🚀 Starting TxAgent App...")
+    agent = init_agent()
+    demo = create_ui(agent)
     demo.queue(api_open=False).launch(
         server_name="0.0.0.0",
         server_port=7860,