coreai_opt.coreai_utils.CompressionGranularity¶

class coreai_opt.coreai_utils.CompressionGranularity(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)[source]¶

Bases: StrEnum

Enum representing the granularity of quantization for Core AI weight compression.

Each member’s string value matches the granularity string accepted by Core AI compression passes.

PER_TENSOR¶: Single set of quantization parameters for the entire tensor.

PER_CHANNEL¶: Separate quantization parameters per individual axis. The targeted axis is pre-defined by the type of operations.

PER_BLOCK¶: Separate quantization parameters per block of axes. The targeted axes are pre-defined by the type of operations.

PER_GROUPED_CHANNEL¶: Separate quantization parameters per group of channels.

__init__(*args, **kwds)¶