File size: 384 Bytes
9891725 |
1 2 3 4 5 6 7 8 9 10 11 |
quant_stage:
quant_modifiers:
QuantizationModifier:
config_groups:
group_0:
input_activations: {dynamic: true, num_bits: 8, observer: memoryless, strategy: token,
symmetric: true, type: float}
targets: [Linear]
weights: {num_bits: 8, observer: mse, strategy: channel, symmetric: true, type: float}
ignore: [lm_head]
|