dmx.compressor.numerical.format.ScaledBlockFloatingPoint

class dmx.compressor.numerical.format.ScaledBlockFloatingPoint(block_format: FixedPoint, scaler_format: FloatingPoint, block_size=64)

This is a scaled block floating point tensor format.

__init__(block_format: FixedPoint, scaler_format: FloatingPoint, block_size=64)

Methods

`__init__`(block_format, scaler_format[, ...])
`cast`(x, block_dim)
`determine_scaler_exponent_bias_from`(x)	Determines scaler_format.bias based on the value of the quantized tensor as a side-effect
`from_shorthand`(sh)

Attributes