Update app.py

app.py

@@ -3,14 +3,13 @@ import pandas as pd
 from datetime import datetime
 from typing import List, Tuple, Dict, Union
 import gradio as gr
-from concurrent.futures import ThreadPoolExecutor
 
 # Constants
 MAX_MODEL_TOKENS = 131072
 MAX_NEW_TOKENS = 4096
 MAX_CHUNK_TOKENS = 8192
 PROMPT_OVERHEAD = 300
-BATCH_SIZE = 3  # NEW: batching chunks to avoid overload
+BATCH_SIZE = 3  # group 3 chunks together
 
 # Paths
 persistent_dir = "/data/hf_cache"
@@ -84,14 +83,14 @@ def init_agent() -> TxAgent:
     agent.init_model()
     return agent
 
-# ⚡ Parallel analyzer with batching
-def analyze_parallel(agent, batch_chunks: List[List[str]], max_workers: int = 3) -> List[str]:
-    results = [None] * len(batch_chunks)
-
-    def worker(index, batch):
+# Serial processing (safe for vLLM)
+def analyze_serial(agent, batch_chunks: List[List[str]]) -> List[str]:
+    results = []
+    for idx, batch in enumerate(batch_chunks):
         prompt = "\n\n".join(build_prompt(chunk) for chunk in batch)
         if estimate_tokens(prompt) > MAX_MODEL_TOKENS:
-            return f"❌ Batch {index+1} too long. Skipped."
+            results.append(f"❌ Batch {idx+1} too long. Skipped.")
+            continue
         response = ""
         try:
             for r in agent.run_gradio_chat(
@@ -111,19 +110,9 @@ def analyze_parallel(agent, batch_chunks: List[List[str]], max_workers: int = 3)
                             response += m.content
                 elif hasattr(r, "content"):
                     response += r.content
-            return clean_response(response)
+            results.append(clean_response(response))
         except Exception as e:
-            return f"❌ Error in batch {index+1}: {str(e)}"
-
-    with ThreadPoolExecutor(max_workers=max_workers) as executor:
-        futures = {executor.submit(worker, idx, batch): idx for idx, batch in enumerate(batch_chunks)}
-        for future in futures:
-            idx = futures[future]
-            try:
-                results[idx] = future.result()
-            except Exception as e:
-                results[idx] = f"❌ Error in batch {idx+1}: {str(e)}"
-
+            results.append(f"❌ Error in batch {idx+1}: {str(e)}")
     gc.collect()
     return results
 
@@ -161,7 +150,7 @@ def process_report(agent, file, messages: List[Dict[str, str]]) -> Tuple[List[Di
         batch_chunks = [chunks[i:i+BATCH_SIZE] for i in range(0, len(chunks), BATCH_SIZE)]
         messages.append({"role": "assistant", "content": f"🔍 Split into {len(batch_chunks)} batches. Analyzing..."})
 
-        chunk_results = analyze_parallel(agent, batch_chunks, max_workers=3)
+        chunk_results = analyze_serial(agent, batch_chunks)
         valid = [res for res in chunk_results if not res.startswith("❌")]
 
         if not valid:
@@ -181,15 +170,14 @@ def process_report(agent, file, messages: List[Dict[str, str]]) -> Tuple[List[Di
         messages.append({"role": "assistant", "content": f"❌ Error: {str(e)}"})
         return messages, None
 
-
 def create_ui(agent):
     with gr.Blocks(css="""
     html, body, .gradio-container {
         background-color: #0e1621;
         color: #e0e0e0;
         font-family: 'Inter', sans-serif;
-        margin: 0;
         padding: 0;
+        margin: 0;
     }
     h2, h3, h4 {
         color: #89b4fa;
@@ -215,7 +203,6 @@ def create_ui(agent):
     }
     .gr-chatbot .message {
         font-size: 16px;
-        line-height: 1.6;
         padding: 12px 16px;
         border-radius: 18px;
         margin: 8px 0;
@@ -242,10 +229,8 @@ def create_ui(agent):
         <h2>📄 CPS: Clinical Patient Support System</h2>
         <p>CPS Assistant helps you analyze and summarize unstructured medical files using AI.</p>
         """)
-
         with gr.Column():
             chatbot = gr.Chatbot(label="CPS Assistant", height=700, type="messages")
-
             upload = gr.File(label="Upload Medical File", file_types=[".xlsx"])
             analyze = gr.Button("🧠 Analyze", variant="primary")
             download = gr.File(label="Download Report", visible=False, interactive=False)
@@ -268,4 +253,3 @@ if __name__ == "__main__":
     except Exception as err:
         print(f"Startup failed: {err}")
         sys.exit(1)
-