Module: tf.keras.quantizers

See Stable See Nightly

DO NOT EDIT.

This file was autogenerated. Do not edit it by hand, since your modifications would be overwritten.

Classes

class AWQConfig: Configuration class for AWQ (Activation-aware Weight Quantization).

class AbsMaxQuantizer

class Float8QuantizationConfig: FP8 quantization config.

class GPTQConfig: Configuration class for the GPTQ (Gradient-based Post-Training Quantization) algorithm.

class Int4QuantizationConfig: Int4 quantization config.

class Int8QuantizationConfig: Int8 quantization config.

class QuantizationConfig: Base class for quantization configs.

class Quantizer

Functions

abs_max_quantize(...): Quantizes the input tensor using the absolute maximum quantization scheme.

abs_max_quantize_grouped_with_zero_point(...): Quantizes a 2D tensor using grouped asymmetric quantization with zero point.

compute_float8_amax_history(...)

compute_float8_scale(...)

deserialize(...): Return a Keras quantizer object via its config.

fake_quant_with_min_max_vars(...): Perform per-tensor or per-channel fake quantization.

get(...): Retrieve a Keras quantizer object via an identifier.

pack_int4(...): Pack an int4 tensor into an int8 tensor with packed nibbles.

quantize_and_dequantize(...)

serialize(...)

unpack_int4(...): Unpack a packed int4 back to an int8 tensor in the range [-8, 7].