.

Dockerfile

@@ -149,6 +149,7 @@ RUN ls -lh /tmp/wheels && \
     python3 -m pip install --no-cache-dir --no-deps /tmp/wheels/vllm-*.whl && \
     python3 -m pip install --no-cache-dir --no-deps /tmp/wheels/sgl_kernel-*.whl && \
     python3 -m pip install --no-cache-dir --no-deps /tmp/wheels/* && \
+    python3 -m pip install --no-cache-dir --upgrade "gradio>=4.44" && \
     python3 -c "from torch.distributed import Backend; print('✅ Runtime torch distributed OK, GLOO =', Backend.GLOO)" && \
     rm -rf /tmp/wheels
 

meta_ui.py

@@ -94,9 +94,13 @@ with gr.Blocks(title="Base 模型测试 UI") as demo:
         test_btn    = gr.Button("🔁 测试 API 可用性")
         test_output = gr.Textbox(label="API 测试结果", interactive=False)
 
+    def chat_with_config(message, history, max_tokens, temperature):
+        return chat(message, history, max_tokens, temperature)
+
     chatbot = gr.ChatInterface(
-        fn=lambda m, h: chat(m, h, max_tokens.value, temperature.value),
-        additional_inputs=[max_tokens, temperature],  # ✅ 添加这一行
+        fn=chat_with_config,
+        additional_inputs=[max_tokens, temperature],
+        type="messages",
         title=None
     )