Update app.py

app.py

@@ -4,7 +4,7 @@ import torch
 import os
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
-# Get model name from secret
+# Load model from Hugging Face secret
 model_name = os.environ.get("MODEL_ID")
 
 tokenizer = AutoTokenizer.from_pretrained(model_name)
@@ -18,16 +18,8 @@ model.eval()
 SYSTEM_PROMPT = "You are an expert biomedical assistant trained to identify randomized controlled trials (RCTs). Include RCTs and exclude non-RCTs."
 
 @spaces.GPU(duration=120)
-def chat_with_llm(message, history):
-    # Start prompt with system role
-    chat_prompt = f"<|system|>\n{SYSTEM_PROMPT.strip()}\n"
-
-    for turn in history:
-        if len(turn) == 2:
-            user_msg, bot_reply = turn
-            chat_prompt += f"<|user|>\n{user_msg.strip()}\n<|assistant|>\n{bot_reply.strip()}\n"
-
-    chat_prompt += f"<|user|>\n{message.strip()}\n<|assistant|>\n"
+def chat_with_llm(message):
+    chat_prompt = f"<|system|>\n{SYSTEM_PROMPT.strip()}\n<|user|>\n{message.strip()}\n<|assistant|>\n"
 
     inputs = tokenizer(chat_prompt, return_tensors="pt", return_token_type_ids=False).to(model.device)
     inputs.pop("token_type_ids", None)
@@ -50,28 +42,26 @@ def chat_with_llm(message, history):
     return reply
 
 with gr.Blocks() as demo:
-    gr.Markdown("## 🧠 RCT Classifier Demonstration")
+    gr.Markdown("## 🧠 RCT Classifier Demonstration (Stateless)")
 
     chatbot = gr.Chatbot()
+
     with gr.Row():
-        title = gr.Textbox(label="Title", placeholder="Enter title")
+        title = gr.Textbox(label="Title", placeholder="Enter article title")
         abstract = gr.Textbox(label="Abstract", placeholder="Enter abstract", lines=6)
 
-    submit = gr.Button("Submit")
-
-    state = gr.State([])
+    submit = gr.Button("Classify")
 
-    def handle_submit(title_text, abstract_text, history):
+    def handle_submit(title_text, abstract_text):
         message = f"Title: {title_text.strip()}\nAbstract: {abstract_text.strip()}"
-        response = chat_with_llm(message, history)
-        history.append((message, response))
-        return history, history
+        response = chat_with_llm(message)
+        return [(message, response)]  # Single-turn memory
 
     submit.click(
         fn=handle_submit,
-        inputs=[title, abstract, state],
-        outputs=[chatbot, state]
+        inputs=[title, abstract],
+        outputs=chatbot
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()