update ui and parameters

app.py

@@ -16,10 +16,10 @@ urllib.request.urlretrieve(
 parser = argparse.ArgumentParser()
 parser.add_argument("-m", "--model", default=DEFAULT_MODEL_PATH, type=Path, help="model path")
 parser.add_argument("--mode", default="chat", type=str, choices=["chat", "generate"], help="inference mode")
-parser.add_argument("-l", "--max_length", default=64, type=int, help="max total length including prompt and output")
+parser.add_argument("-l", "--max_new_tokens", default=64, type=int, help="max total output tokens")
 parser.add_argument("-c", "--max_context_length", default=1024, type=int, help="max context length")
 parser.add_argument("--top_k", default=40, type=int, help="top-k sampling")
-parser.add_argument("--top_p", default=0.7, type=float, help="top-p sampling")
+parser.add_argument("--top_p", default=0.75, type=float, help="top-p sampling")
 parser.add_argument("--temp", default=0.5, type=float, help="temperature")
 parser.add_argument("--repeat_penalty", default=1.0, type=float, help="penalize repeat sequence of tokens")
 parser.add_argument("-t", "--threads", default=0, type=int, help="number of threads for inference")
@@ -36,13 +36,13 @@ def postprocess(text):
     return text
 
 
-def predict(input, chatbot, max_length, top_p, temperature, messages):
+def predict(input, chatbot, max_new_tokens, top_p, temperature, messages):
     chatbot.append((postprocess(input), ""))
     messages.append(chatglm_cpp.ChatMessage(role="user", content=input))
     full_messages = [system_message] + messages
 
     generation_kwargs = dict(
-        max_length=max_length,
+        max_new_tokens=max_new_tokens,
         max_context_length=args.max_context_length,
         do_sample=temperature > 0,
         top_k=args.top_k,
@@ -83,13 +83,13 @@ title = """
 with gr.Blocks() as demo:
     gr.HTML(title)
 
-    chatbot = gr.Chatbot()
+    chatbot = gr.Chatbot().style(height=300)
     with gr.Row():
         with gr.Column(scale=4):
             user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=8)
             submitBtn = gr.Button("Submit", variant="primary")
         with gr.Column(scale=1):
-            max_length = gr.Slider(0, 512, value=args.max_length, step=1.0, label="Maximum Length", interactive=True)
+            max_new_tokens = gr.Slider(0, 512, value=args.max_new_tokens, step=1.0, label="Maximum output tokens", interactive=True)
             top_p = gr.Slider(0, 1, value=args.top_p, step=0.01, label="Top P", interactive=True)
             temperature = gr.Slider(0, 1, value=args.temp, step=0.01, label="Temperature", interactive=True)
             emptyBtn = gr.Button("Clear History")
@@ -98,7 +98,7 @@ with gr.Blocks() as demo:
 
     submitBtn.click(
         predict,
-        [user_input, chatbot, max_length, top_p, temperature, messages],
+        [user_input, chatbot, max_new_tokens, top_p, temperature, messages],
         [chatbot, messages],
         show_progress=True,
     )