afe.core.mixed_precision.config

Functions

get_core_config(...)	Get mct's core config, with details about error methods, bias correction, etc.
get_model_target_platform_capabilities(...)	Get mct's target platform capabilities that are used for quantization like quantization method,

Module Contents

afe.core.mixed_precision.config.get_core_config() → model_compression_toolkit.core.CoreConfig

Get mct’s core config, with details about error methods, bias correction, etc. TODO: Currently a fixed config, needs to be configurable based on user’s choices from AFE specs.

Returns:

core_config. CoreConfig with quantization specific details.

afe.core.mixed_precision.config.get_model_target_platform_capabilities(activation_n_bits: int) → model_compression_toolkit.target_platform.TargetPlatformCapabilities

Get mct’s target platform capabilities that are used for quantization like quantization method, per channel weights, layer fusing etc.

TODO: Currently fixed specifications, but need to modified based on quantization params and

AFE’s layer fusions.

Parameters:

activation_n_bits – Bit setting to quantize to.

Returns:

pytorch_tpc. Target Platform Capability object with quantization details.