FlatInt8AT

python3 flat_at_int8_generate_model.py --cores <core_num> -d <dim> -c <code_num> -p <process_id> --soc-version <soc_version> -t <npu_type>

<core_num>: number of AI Cores of the Ascend AI Processor. The default value is 8.

<dim>: input feature vector dimension. The default value is 256.

<code_num>: number of base library features to be compared with the input features. The default value is 16384.

<process_id>: ID of the process for multi-process scheduling of operators generated in batches. The default value is 0, and you do not need to set this parameter.

<soc_version>: Ascend AI Processor model. The default value is Ascend310P3. You do not need to set it.

<npu_type>: hardware form. Currently, it supports the Atlas inference product. The default value is 310P and does not need to be set.

--help | -h: help information.

Run the command to obtain a group of operator model files.

FlatInt8AT reduces the time required by train, add, and update of IVFSQT in the Atlas inference product.

• dim ∈ {256}
• code_num ∈ {1024, 2048, 4096, 8192, 16384, 32768}
• soc_version ∈ {Ascend310P3}