tf.keras.quantizers.Float8QuantizationConfig

See Stable See Nightly

View source on GitHub

FP8 quantization config.

Inherits From: QuantizationConfig

tf.keras.quantizers.Float8QuantizationConfig()

FP8 mixed-precision training does not support user defined quantizers. This config is only used to indicate that FP8 mixed-precision training should be used.

Methods

activation_quantizer_or_default

View source

@staticmethod
activation_quantizer_or_default(
    config, default
)

from_config

View source

@classmethod
from_config(
    config
)

get_config

View source

get_config()

weight_quantizer_or_default

View source

@staticmethod
weight_quantizer_or_default(
    config, default
)