Update app.py

app.py

@@ -16,14 +16,12 @@ def generate_response(prompt_template, **kwargs):
         top_p=0.8,
         stream=True
     )
-    response = ""
     for chunk in stream:
-        response += chunk.choices[0].delta.content
-        yield response  # Yield incrementally for streaming
+        yield chunk.choices[0].delta.content  # Yield chunks as they are generated
 
-# Wrapper function for Gradio compatibility
-def stream_response(prompt_template, **kwargs):
-    return generate_response(prompt_template, **kwargs)
+# Wrapper function to combine generator output
+def get_response(prompt_template, **kwargs):
+    return "".join(generate_response(prompt_template, **kwargs))
 
 # Gradio app interface
 with gr.Blocks() as app:
@@ -51,7 +49,7 @@ with gr.Blocks() as app:
             output = gr.Markdown(label="Output", elem_id="latex-output")
             # Link button to the response wrapper
             button.click(
-                fn=lambda *args: stream_response(prompt_template, **dict(zip([inp["key"] for inp in inputs], args))),
+                fn=lambda *args: get_response(prompt_template, **dict(zip([inp["key"] for inp in inputs], args))),
                 inputs=input_fields,
                 outputs=output,
                 api_name=f"/{tab_name.lower().replace(' ', '_')}_execute"