ptuning微调运行后进度条为0，一动不动，求帮助解决

[HHCOlll]

  train_file: finetune_demo/data/train.json
  val_file: finetune_demo/data/validation.json
  test_file: finetune_demo/data/test.json
  num_proc: 1
max_input_length: 1024
max_output_length: 512
training_args:
  # see `transformers.Seq2SeqTrainingArguments`
  output_dir: ./output
  max_steps: 18000
  # settings for data loading
  per_device_train_batch_size: 1
  dataloader_num_workers: 1
  remove_unused_columns: false
  # settings for saving checkpoints
  save_strategy: steps
  save_steps: 3000
  # settings for logging
  log_level: info
  logging_strategy: steps
  logging_steps: 50
  # settings for evaluation
  per_device_eval_batch_size: 1
  evaluation_strategy: steps
  eval_steps: 500
  # settings for optimizer
  # adam_epsilon: 1e-6
  # uncomment the following line to detect nan or inf values
  # debug: underflow_overflow
  predict_with_generate: true
  # see `transformers.GenerationConfig`
  generation_config:
    max_new_tokens: 512
  # set your absolute deepspeed path here
  #deepspeed: ds_zero_3.json
  use_cpu: false
peft_config:
  peft_type: PREFIX_TUNING
  task_type: CAUSAL_LM
  num_virtual_tokens: 128

我的参数是上面这样设置的，运行完是下面这样的

[zRzRzRzRzRzRzR]

你的配置是啥

[HHCOlll]

你好，是A100，64g内存，我现在用初始的demo参数和那个广告训练数据集进行训练依然是训练很慢，我用二代模型就不会

[liujingmao]

可能是你的pytorch没有安装GPU版本的，我遇到过这样的问题，安装对pytorch也是要注意的

[HHCOlll]

感谢，解决了

[mochaphev]

请问你是如何解决的，我也遇到了同样的问题

[HHCOlll]

你好，我是下载了gpu版本的pytorch就好了