AttributeError: 'Tab' object has no attribute 'constructor_args'

[xiezhipeng-git]

Describe the bug
What the bug is, and how to reproduce, better with screenshots(描述bug以及复现过程，最好有截图)
通过webUi 启动
vscode lauch.json

{
            "name": "swift web-ui",
            "type": "debugpy",
            "request": "launch",
            "program": "${workspaceRoot}/swift/cli/web_ui.py",
            // "program": "${workspaceRoot}swift/cli/main.py",
            "args": [
                        // "webui"
                    ],
            "justMyCode": false,
            "console": "integratedTerminal",
            "cwd": "${workspaceFolder}",
            "env": {
                "PYTHONPATH": "${workspaceFolder}${pathSeparator}${env:PYTHONPATH}",
            }
        }

Your hardware and system info
Write your system info like CUDA version/system/GPU/torch version here(在这里给出硬件信息和系统信息，如CUDA版本，系统，GPU型号和torch版本等)
Name: torch
Version: 2.5.1
GPU 4090
CUDA Build cuda_12.6.r12.6/compiler.34841621_0
system windows 10 wsl Ubuntun24.04
Additional context
Add any other context about the problem here(在这里补充其他信息)
最新的github main 版本。另外如果gradio 版本高一些有其他错误

发生异常: AttributeError       (note: full exception trace is shown but execution is paused at: _run_module_as_main)
'Tab' object has no attribute 'constructor_args'
  File "/mnt/d/my/work/LLM/ai_train/ms-swift/swift/ui/base.py", line 63, in wrapper
    self.constructor_args.update(kwargs)
    ^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/d/my/work/LLM/ai_train/ms-swift/swift/ui/llm_train/llm_train.py", line 208, in do_build_ui
    with gr.TabItem(elem_id='llm_train', label=''):
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/d/my/work/LLM/ai_train/ms-swift/swift/ui/base.py", line 120, in build_ui
    cls.do_build_ui(base_tab)
  File "/mnt/d/my/work/LLM/ai_train/ms-swift/swift/ui/app.py", line 64, in run
    LLMTrain.build_ui(LLMTrain)
  File "/mnt/d/my/work/LLM/ai_train/ms-swift/swift/llm/base.py", line 46, in main
    result = self.run()
             ^^^^^^^^^^
  File "/mnt/d/my/work/LLM/ai_train/ms-swift/swift/ui/app.py", line 92, in webui_main
    return SwiftWebUI(args).main()
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/d/my/work/LLM/ai_train/ms-swift/swift/cli/web_ui.py", line 5, in <module>
    webui_main()
  File "/root/anaconda3/lib/python3.12/runpy.py", line 88, in _run_code
    exec(code, run_globals)
  File "/root/anaconda3/lib/python3.12/runpy.py", line 198, in _run_module_as_main (Current frame)
    return _run_code(code, main_globals, None,
AttributeError: 'Tab' object has no attribute 'constructor_args'

看起来 和 Gradio 版本有关。当前使用的版本是3.40.0
并且升级后没有这个问题。但是有其他问题
#2782

Traceback (most recent call last):
  File "/root/anaconda3/lib/python3.12/site-packages/gradio/queueing.py", line 625, in process_events
    response = await route_utils.call_process_api(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/anaconda3/lib/python3.12/site-packages/gradio/route_utils.py", line 322, in call_process_api
    output = await app.get_blocks().process_api(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/anaconda3/lib/python3.12/site-packages/gradio/blocks.py", line 2040, in process_api
    inputs = await self.preprocess_data(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/anaconda3/lib/python3.12/site-packages/gradio/blocks.py", line 1735, in preprocess_data
    processed_input.append(block.preprocess(inputs_cached))
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/anaconda3/lib/python3.12/site-packages/gradio/components/dropdown.py", line 198, in preprocess
    raise Error(
gradio.exceptions.Error: "Value: {'__type__': 'update'} (type: <class 'dict'>) is not in the list of choices: ['qwen', 'qwen2', 'qwen2_5', 'qwen2_5_math', 'qwen2_moe', 'qwq', 'codefuse_qwen', 'modelscope_agent', 'marco_o1', 'llama', 'llama3', 'llama3_1', 'llama3_2', 'reflection', 'megrez', 'yi', 'yi_coder', 'sus', 'codefuse_codellama', 'mengzi3', 'ziya', 'numina', 'atom', 'chatglm2', 'chatglm3', 'glm4', 'glm_edge', 'codefuse_codegeex2', 'codegeex4', 'longwriter_llama3_1', 'internlm', 'internlm2', 'internlm3', 'deepseek', 'deepseek_moe', 'deepseek_v2', 'deepseek_v2_5', 'deepseek_r1', 'deepseek_r1_distill', 'openbuddy_llama', 'openbuddy_llama3', 'openbuddy_mistral', 'openbuddy_mixtral', 'baichuan', 'baichuan2', 'minicpm', 'minicpm_chatml', 'minicpm3', 'minicpm_moe', 'telechat', 'telechat2', 'mistral', 'zephyr', 'mixtral', 'mistral_nemo', 'wizardlm2', 'wizardlm2_moe', 'phi2', 'phi3_small', 'phi3', 'phi3_moe', 'phi4', 'minimax', 'gemma', 'gemma2', 'skywork', 'skywork_o1', 'yuan2', 'orion', 'xverse', 'xverse_moe', 'seggpt', 'bluelm', 'c4ai', 'dbrx', 'grok', 'mamba', 'polylm', 'aya', 'qwen_vl', 'qwen_audio', 'qwen2_vl', 'qwen2_audio', 'qvq', 'ovis1_6', 'ovis1_6_llama3', 'glm4v', 'glm_edge_v', 'cogvlm', 'cogagent_vqa', 'cogagent_chat', 'cogvlm2', 'cogvlm2_video', 'internvl', 'internvl_phi3', 'internvl2', 'internvl2_phi3', 'internvl2_5', 'xcomposer2', 'xcomposer2_4khd', 'xcomposer2_5', 'xcomposer2_5_ol_audio', 'llama3_2_vision', 'llama3_1_omni', 'llava1_5_hf', 'llava1_6_mistral_hf', 'llava1_6_vicuna_hf', 'llava1_6_yi_hf', 'llama3_llava_next_hf', 'llava_next_qwen_hf', 'llava_next_video_hf', 'llava_next_video_yi_hf', 'llava_onevision_hf', 'yi_vl', 'llava_llama3_1_hf', 'llava_llama3_hf', 'llava1_6_mistral', 'llava1_6_yi', 'llava_next_qwen', 'llama3_llava_next', 'deepseek_vl', 'deepseek_vl2', 'deepseek_janus', 'minicpmv', 'minicpmv2_5', 'minicpmv2_6', 'minicpmo2_6', 'minimax_vl', 'mplug_owl2', 'mplug_owl2_1', 'mplug_owl3', 'mplug_owl3_241101', 'doc_owl2', 'emu3_gen', 'emu3_chat', 'got_ocr2', 'phi3_vision', 'florence', 'idefics3', 'paligemma', 'molmo', 'molmoe', 'pixtral', 'megrez_omni', 'valley', 'modern_bert', 'bert', 'internlm2_reward', 'qwen2_reward', 'qwen2_5_prm', 'qwen2_5_math_reward', 'llama3_2_reward', 'gemma_reward']"

@Jintao-Huang

[Jintao-Huang]

@tastelikefeet

[tastelikefeet]

请使用gradio>=3.50.2
第二个错误的复现步骤能给一下吗

[xiezhipeng-git]

@tastelikefeet
和第一个错误的复现步骤基本一样。只需要升级gradio
然后参数随便选，接着点击生成预览命令 就有这个错误了。还没开始训练

[xiezhipeng-git]

@tastelikefeet
更新github代码

[INFO:swift] output_dir: output/Qwen2.5-7B-Instruct/v0-20250204-215033
[INFO:swift] Run training: CUDA_VISIBLE_DEVICES=0  nohup swift sft --torch_dtype 'bfloat16' --model 'Qwen/Qwen2.5-7B-Instruct' --model_type 'qwen2_5' --template 'qwen2_5' --system 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' --dataset 'AI-ModelScope/alpaca-gpt4-data-zh' --max_length '1024' --init_weights 'True' --learning_rate '1e-4' --attn_impl 'flash_attn' --gradient_accumulation_steps '16' --eval_steps '500'  --add_version False --output_dir /mnt/d/my/work/LLM/ai_train/ms-swift/output/Qwen2.5-7B-Instruct/v0-20250204-215033 --logging_dir /mnt/d/my/work/LLM/ai_train/ms-swift/output/Qwen2.5-7B-Instruct/v0-20250204-215033/runs --ignore_args_error True > /mnt/d/my/work/LLM/ai_train/ms-swift/output/Qwen2.5-7B-Instruct/v0-20250204-215033/runs/run.log 2>&1 &

这样的信息代表着生成的运行命令不会报错了是吧。
但是生成运行命令为什么不能像llamafactory 那样。直接展示在web界面。而是在控制台呢？并且应该把训练按钮和生成运行命令分开吧。
然后展示在web界面选择保存为文件，类似llamafactory 。
接着，如果在增加
#2994
然后统一下生成的sh命令的格式。去掉特殊格式，比如目前的dpo.sh的特殊命令。形成生成预览，保存文件，接着直接使用保存的文件运行训练。整体预览命令功能就闭环了