Shubhra Pandit
Update model files - no router quantization, actorder static
4966942
raw
history blame contribute delete
353 Bytes
quant_stage:
quant_modifiers:
GPTQModifier:
sequential_update: true
dampening_frac: 0.01
ignore: [lm_head, 're:.*router*']
config_groups:
group_0:
weights: {num_bits: 4, type: int, symmetric: true, strategy: group, group_size: 128,
actorder: weight, observer: mse}
targets: [Linear]