部署NLP模型遇FileNotFoundError，pickle与.h5保存均报错求助

报错信息 #

FileNotFoundError: Unsuccessful TensorSliceReader constructor: Failed to find any matching files for ram://a603e930-4fda-4105-8554-7af5e5fc02f5/variables/variables
You may be trying to load on a different device from the computational device. Consider setting the experimental_io_device option in tf.saved_model.LoadOptions to the io_device such as '/job:localhost'

问题背景 #

先后使用pickle、.h5格式保存NLP模型，加载时均触发上述错误。

解决方法 #

• 加载模型时指定experimental_io_device参数，强制使用本地设备读取文件：

  import tensorflow as tf
  
  load_options = tf.saved_model.LoadOptions(experimental_io_device='/job:localhost')
  model = tf.keras.models.load_model('你的模型文件路径.h5', options=load_options)
  

• 重新完整保存模型：确保模型保存过程未中断，执行以下代码重新生成.h5文件：

  model.save('完整的模型保存路径.h5', save_format='h5')
  

• 明确指定加载设备：若训练与加载设备不一致（如训练用GPU，加载用CPU），添加设备上下文：

  import tensorflow as tf
  
  load_options = tf.saved_model.LoadOptions(experimental_io_device='/job:localhost')
  with tf.device('/CPU:0'):
      model = tf.keras.models.load_model('你的模型文件路径.h5', options=load_options)
  

内容的提问来源于stack exchange，提问作者Karthik Bhandary