app.py

@@ -42,7 +42,7 @@ if has_cuda:
         )
 else:
     model = (
-        AutoModel.from_pretrained(model_name, trust_remote_code=True).half().float()
+        AutoModel.from_pretrained(model_name, trust_remote_code=True).float()
     )  # .float() .half().float(): must use float for cpu
 
 model = model.eval()
@@ -94,9 +94,10 @@ with gr.Blocks(theme=theme) as block:
                 clear = gr.ClearButton([msg, chatbot])
 
     # do not clear prompt
-    msg.submit(lambda x, y: (x,) + respond(x, y)[1:], [msg, chatbot], [msg, chatbot])
+    btn.submit(respond, [msg, chatbot], [msg, chatbot])
+
+    msg.click(lambda x, y: ("",) + respond(x, y)[1:], [msg, chatbot], [msg, chatbot])
 
-    btn.click(respond, [msg, chatbot], [msg, chatbot])
 
     with gr.Accordion("Example inputs", open=True):
         etext = """In America, where cars are an important part of the national psyche, a decade ago people had suddenly started to drive less, which had not happened since the oil shocks of the 1970s. """