Documentation

Result Files

For single-operator comparison, the absolute error and relative error in data of a specified operator are compared by default. The comparison result file is generated based on the input or output data of the operator and the configured output file format, as shown in Table 1.

Table 1 Result file formats

Result File Output Condition

Result File

Compare the input data and generate the first 20 items of data by default.

--input_tensor

Single-operator comparison result (summary): {op_name}_input_{index}_summary.txt

Single-operator comparison result (complete): {op_name}_input_{index}_{file_index}.csv

Single-operator comparison result with top 20 items:

  • Absolute error: {op_name}_input_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_input_{index}_relative_error_topn.csv

Compare the output data and generate the first 20 items of data by default.

--output_tensor

Single-operator comparison result (summary): {op_name}_output_{index}_summary.txt

Single-operator comparison result (complete): {op_name}_output_{index}_{file_index}.csv

Single-operator comparison result with top 20 items:

  • Absolute error: {op_name}_output_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_output_{index}_relative_error_topn.csv

Compare the input data and generate the first n items of data.

--input_tensor, --topn

Single-operator comparison result (summary): {op_name}_input_{index}_summary.txt

Single-operator comparison result (complete): {op_name}_input_{index}_{file_index}.csv

Single-operator comparison result with top n items:

  • Absolute error: {op_name}_input_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_input_{index}_relative_error_topn.csv

Compare the output data and generate the first n items of data.

--output_tensor, --topn

Single-operator comparison result (summary): {op_name}_output_{index}_summary.txt

Single-operator comparison result (complete): {op_name}_output_{index}_{file_index}.csv

Single-operator comparison result with top n items:

  • Absolute error: {op_name}_output_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_output_{index}_relative_error_topn.csv

Compare the input data and generate the first 20 items of data by default. Complete comparison data is not generated.

--input_tensor, --ignore_single_op_result

Single-operator comparison result (summary): {op_name}_input_{index}_summary.txt

Single-operator comparison result with top 20 items:

  • Absolute error: {op_name}_input_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_input_{index}_relative_error_topn.csv

Compare the output data and generate the first 20 items of data by default. Complete comparison data is not generated.

--output_tensor, --ignore_single_op_result

Single-operator comparison result (summary): {op_name}_output_{index}_summary.txt

Single-operator comparison result with top 20 items:

  • Absolute error: {op_name}_output_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_output_{index}_relative_error_topn.csv

Compare the input data, configure the maximum number of files contained in a single .csv file, and generate the first 20 items of data by default.

--input_tensor, --max_line

Single-operator comparison result (summary): {op_name}_input_{index}_summary.txt

Single-operator comparison result (split): {op_name}_input_{index}_{file_index}.csv

Single-operator comparison result with top 20 items:

  • Absolute error: {op_name}_input_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_input_{index}_relative_error_topn.csv

Compare the output data, configure the maximum number of files contained in a single .csv file, and generate the first 20 items of data by default.

--output_tensor, --max_line

Single-operator comparison result (summary): {op_name}_output_{index}_summary.txt

Single-operator comparison result (split): {op_name}_output_{index}_{file_index}.csv

Single-operator comparison result with top 20 items:

  • Absolute error: {op_name}_output_{index}_absolute_error_topn.csv
  • Relative error: {op_name}_output_{index}_relative_error_topn.csv

Note 1: If the comparison operator name {op_name} is too long, it is automatically converted to the name of the dump file where the operator is located. The mapping simple_op_mapping.csv between the generated result files and the actual result files (before operator name conversion) is added.

Note 2: file_index indicates the ID of the output .csv file, whose default value is 0. When --max_line is configured, its value is "(n-1) x max_line", where n is a positive integer starting from 1.

Note 3: For details about the preceding parameters, see Command Syntax.