Error Handling

Infery Exceptions

Infery includes error-handling logically.

In this page we will go through their meaning.

General Exceptions

  • InferyError - A base class for all the of infery's exceptions.

Inference (Runtime) Exceptions

  • InputTypeError - Inputs are passed in wrong data structure (e.g. torch.Tensor instead of List).
  • ModelFrameworkRuntimeError - An error that occurred in the model's framework runtime, after the model was loaded.
  • ForwardPassNotImplemented - A pure forward-pass method is not implemented for the inferencer, thus it cannot be used.

Usage Exceptions

  • DTypeError - Mismatch between expected dtype and received dtype.
  • MaxBatchSizeExceededError - In dynamic batch size models, Specified batch size exceeds maximum value allowed by model.
  • MappingKeyError - Unrecognized key or attribute was passed to a mapping data structure (dict, Enum, Schema, etc.)
  • ModelFrameworkDetectionError - The framework type could not be automatically inferred from the checkpoint. Please specify the framework type explicitly, by passing 'framework_type=...' to infery.load.
  • EmptyInputValue - A required value was not provided.

Model Loading Exceptions

  • FrameworkNotSupportedError - The specified framework is not supported. Please refer to FrameworkType enum for valid values.
  • CheckpointFileDoesNotExistError - The specified checkpoint file does not exist at the specified path.
  • InvalidInputDimensions - The specified input dimensions is not a valid tuple.
  • InvalidStaticBatchSize - The specified static batch size is not a valid int.
  • InvalidInferenceHardware - The specified inference hardware is not valid. Please refer to InferenceHardware enum for valid values.
  • InvalidLoggingLevel - The specified logging level is not valid. Please select from the logging package's standard logging level values.
  • InputDimensionsNotSpecifiedError - The model's input dimensions were not specified, but they are required for benchmarks inputs generation.
    Please specify the input dimensions via 'input_dims' argument.
  • FailedToLoadCheckpointError - The checkpoint failed to load for any reason.
  • CorruptedModelCompressionError - Failed to unpack the saved model zip/tar file, corrupted zip/tar.
  • CorruptedSavedModelZipError - Failed to unpack the saved model zip file, corrupted zip.
  • BadSavedModelZipError - Failed to load the saved model from the provided zip file, it may be missing.
  • MissingFileModelError - Failed to load the saved model from the provided zip/tar file, the model is missing.
  • ModelFrameworkNotLoadedError - An internal error indicating an import is missing for running inference for the model's framework.
  • ModelValueError - An error that occurred in the model's initialization/preperation, before or while the model was loaded.