IVFSQT

To reduce the time required for train and add functions, generate the FlatAT operator. <dim> must be the same as <dim_in>, and <code_num> must be the same as <coarse_centroid_num>.

The IVFast solution is the same as the IVFSQT solution.

Usage

python3 ivfsqt_generate_model.py --cores <core_num> -d <dim_in> -r <compress_ratio> -c <coarse_centroid_num> -p <process_id>

Parameter

<dim_in>: dimension of the input feature vector.

<compress_ratio>: ratio of the input dimension to the output dimension.

<coarse_centroid_num>: number of L1 cluster centroids.

<core_num>: number of AI Cores of the Ascend AI Processor. The default value is 2.

<process_id>: ID of the process for batch generating operator multi-process scheduling tasks.

Description

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

Restrictions

  • dimIn ∈ {256}
  • dimOut ∈ {64}
  • coarse_centroid_num ∈ {1024, 2048, 4096, 8192, 16384, 32768}
  • <dim_in> can be exactly divided by <compress_ratio>.