33 lines
880 B
YAML
33 lines
880 B
YAML
|
default_stage:
|
||
|
|
default_modifiers:
|
||
|
|
QuantizationModifier:
|
||
|
|
config_groups:
|
||
|
|
group_0:
|
||
|
|
targets: [Linear]
|
||
|
|
weights:
|
||
|
|
num_bits: 4
|
||
|
|
type: float
|
||
|
|
symmetric: true
|
||
|
|
group_size: 16
|
||
|
|
strategy: tensor_group
|
||
|
|
block_structure: null
|
||
|
|
dynamic: false
|
||
|
|
actorder: null
|
||
|
|
observer: mse
|
||
|
|
observer_kwargs: {}
|
||
|
|
input_activations:
|
||
|
|
num_bits: 4
|
||
|
|
type: float
|
||
|
|
symmetric: true
|
||
|
|
group_size: 16
|
||
|
|
strategy: tensor_group
|
||
|
|
block_structure: null
|
||
|
|
dynamic: local
|
||
|
|
actorder: null
|
||
|
|
observer: minmax
|
||
|
|
observer_kwargs: {}
|
||
|
|
output_activations: null
|
||
|
|
format: null
|
||
|
|
targets: [Linear]
|
||
|
|
ignore: ['re:.*lm_head.*', 're:.*mlp.gate$']
|