dmx.compressor.numerical.format.BlockFloatingPoint

class dmx.compressor.numerical.format.BlockFloatingPoint(precision=8, block_size=64, symmetric=True, rounding='nearest')

This is a block floating point format simulated in FP32, using QPyTorch.

__init__(precision=8, block_size=64, symmetric=True, rounding='nearest')

Methods

__init__([precision, block_size, symmetric, ...])
cast(x, block_dim)
from_shorthand(sh)
make_mantissa_asymmetric(dmx_result, fp32_inp)

Attributes

bfp_id
bit_precision
blocked
bytes_per_elem