dmx.compressor.numerical.format.MXFP

class dmx.compressor.numerical.format.MXFP(element_format: FloatingPoint, block_size=32)

This is a MXFP tensor format.

__init__(element_format: FloatingPoint, block_size=32)

Methods

__init__(element_format[, block_size])
cast(x, block_dim)
from_shorthand(sh)

Attributes

bfp_id
bit_precision
blocked
bytes_per_elem