core.enums#

Module Contents#

Classes#

ModelType

Model type.

Fp8Recipe

FP8 recipe names: delayed, tensorwise, mxfp8, blockwise, custom.

Fp4Recipe

FP4 recipe names: nvfp4, custom.

API#

class core.enums.ModelType(*args, **kwds)#

Bases: enum.Enum

Model type.

Initialization

encoder_or_decoder#

1

retro_encoder#

2

retro_decoder#

3

property encoder_and_decoder#

Deprecated property - use encoder_or_decoder instead.

class core.enums.Fp8Recipe#

Bases: str, enum.Enum

FP8 recipe names: delayed, tensorwise, mxfp8, blockwise, custom.

Initialization

Initialize self. See help(type(self)) for accurate signature.

delayed#

‘delayed’

tensorwise#

‘tensorwise’

mxfp8#

‘mxfp8’

blockwise#

‘blockwise’

custom#

‘custom’

class core.enums.Fp4Recipe#

Bases: str, enum.Enum

FP4 recipe names: nvfp4, custom.

Initialization

Initialize self. See help(type(self)) for accurate signature.

nvfp4#

‘nvfp4’

custom#

‘custom’