dmx.compressor.numerical.format.MXINT

class dmx.compressor.numerical.format.MXINT(precision=8, block_size=32)

This is a MXINT tensor format.

__init__(precision=8, block_size=32)

Methods

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

Attributes

bfp_id
bit_precision
blocked
bytes_per_elem