Update app.py

app.py

@@ -166,27 +166,27 @@ def http_bot(state, model_selector, temperature, top_p, max_new_tokens, request:
     yield (state, state.to_gradio_chatbot()) + (disable_btn,) * 5
 
 
-    # try:
+    try:
     # Stream output
-    response = chat.generate_stream_gate(pload)
-    for chunk in response:
-        if chunk:
-            data = json.loads(chunk.decode())
-            if data["error_code"] == 0:
-                output = data["text"][len(prompt):].strip()
-                state.messages[-1][-1] = output + "▌"
-                yield (state, state.to_gradio_chatbot()) + (disable_btn,) * 5
-            else:
-                output = data["text"] + f" (error_code: {data['error_code']})"
-                state.messages[-1][-1] = output
-                yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
-                return
-            time.sleep(0.03)
-    # except Exception:
-    #     gr.Warning(server_error_msg)
-    #     state.messages[-1][-1] = server_error_msg
-    #     yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
-    #     return
+        response = chat.generate_stream_gate(pload)
+        for chunk in response:
+            if chunk:
+                data = json.loads(chunk.decode())
+                if data["error_code"] == 0:
+                    output = data["text"][len(prompt):].strip()
+                    state.messages[-1][-1] = output + "▌"
+                    yield (state, state.to_gradio_chatbot()) + (disable_btn,) * 5
+                else:
+                    output = data["text"] + f" (error_code: {data['error_code']})"
+                    state.messages[-1][-1] = output
+                    yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
+                    return
+                time.sleep(0.03)
+    except Exception:
+        gr.Warning(server_error_msg)
+        state.messages[-1][-1] = server_error_msg
+        yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
+        return
 
     state.messages[-1][-1] = state.messages[-1][-1][:-1]
     yield (state, state.to_gradio_chatbot()) + (enable_btn,) * 5
@@ -233,7 +233,7 @@ def build_demo(embed_mode):
             with gr.Column(scale=3):
                 with gr.Row(elem_id="model_selector_row"):
                     model_selector = gr.Dropdown(
-                        choices=models,
+                        choices=[model + "-4bit" for model in models],
                         value=models[0]+"-4bit" if len(models) > 0 else "",
                         interactive=True,
                         show_label=False,