Documentation

Dump File Naming Conventions

The current version supports multiple comparison approaches. Keep the following naming conventions in mind when creating dump files and .npy files.

Table 1 Naming conventions of dump files

Data Type

Naming Format

Remarks

Dump data file of the non-quantized offline model running on the Ascend AI Processor

Currently, the following file name formats are supported:

  • {op_type}.{op_name}.{task_id}.{stream_id}.{timestamp}
  • {op_type}.{op_name_lxsliceX}.({stream_id}.){task_id}.{timestamp}.{task_type}.{context_id}.{thread_id}.{device_id}
  • {op_type}.{op_name}.({stream_id}.){task_id}.{timestamp}.{task_type}.{context_id}.{thread_id}.{device_id}

The name formats are described as follows:

  • The names of op_type (operator type) and op_name (operator name) must comply with the A-Za-z0-9_- regular expression rule.
  • timestamp is a 16-bit timestamp.
  • Each of the following parameters consists of digits 0-9: task_id (task ID), stream_id (stream ID), output_index (Nth output), task_type (task type), context_id (context ID), thread_id (thread ID), device_id (device ID)

Dump data file of the quantized offline model running on the Ascend AI Processor

.npy file (CaffeeTensorFlow/ONNX)

{op_name}.{output_index}.{timestamp}.npy