Update app.py

app.py

@@ -27,9 +27,9 @@ def stream_response(title_text, abstract_text):
 
     generated = inputs["input_ids"]
     past_key_values = None
-    response_text = ""
 
-    for _ in range(1024):  # limit max length
+    reply = ""
+    for _ in range(1024):
         outputs = model(input_ids=generated, past_key_values=past_key_values, use_cache=True)
         next_token_logits = outputs.logits[:, -1, :]
         next_token = torch.argmax(next_token_logits, dim=-1, keepdim=True)
@@ -41,28 +41,33 @@ def stream_response(title_text, abstract_text):
         decoded_output = tokenizer.decode(generated[0], skip_special_tokens=True)
 
         if "<|assistant|>" in decoded_output:
-            response_text = decoded_output.split("<|assistant|>")[-1].strip()
-            yield response_text
+            reply = decoded_output.split("<|assistant|>")[-1].strip()
+            yield [(user_input, reply)]
 
 with gr.Blocks() as demo:
-    gr.Markdown("## 🧠 RCT Classifier Demonstration (Streaming Enabled)")
+    gr.Markdown("## 🧠 RCT Classifier Demonstration (Streaming, Stateless)")
+
+    chatbot = gr.Chatbot(label="Model Response", type="tuple")
 
-    chatbot = gr.Chatbot()
     with gr.Row():
         title = gr.Textbox(label="Title", placeholder="Enter title")
         abstract = gr.Textbox(label="Abstract", placeholder="Enter abstract", lines=6)
+
     submit = gr.Button("Classify")
 
     def stream_chat(title_text, abstract_text):
-        user_message = f"Title: {title_text.strip()}\nAbstract: {abstract_text.strip()}"
-        yield (user_message, "")  # show user message
-        response_stream = stream_response(title_text, abstract_text)
-        collected = ""
-        for partial in response_stream:
-            collected = partial
-            yield (user_message, collected)
-
-    submit.stream(fn=stream_chat, inputs=[title, abstract], outputs=chatbot)
+        user_input = f"Title: {title_text.strip()}\nAbstract: {abstract_text.strip()}"
+        yield [(user_input, "")]  # show initial empty response
+
+        response_generator = stream_response(title_text, abstract_text)
+        for update in response_generator:
+            yield update
+
+    submit.click(
+        fn=stream_chat,
+        inputs=[title, abstract],
+        outputs=chatbot
+    )
 
 if __name__ == "__main__":
     demo.launch()