You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Apr 28, 2024. It is now read-only.
code: result_dora = (mag_norm_scale - 1) * (F.linear(x, transpose(weight, self.fan_in_fan_out)) ) + mag_norm_scale * lora_B(lora_A(x)) * scaling
Question: what is the effect of (mag_norm_scale - 1) and mag_norm_scale ? And, result_dora can't equals the F.linear(x, transpose(weight, self.fan_in_fan_out)) in the Initializing stage due to the parameter "mag_norm_scale - 1"
The text was updated successfully, but these errors were encountered:
Note that (mag_norm_scale - 1) * (F.linear(x, transpose(weight, self.fan_in_fan_out))) must be included to properly apply dropout; otherwise, the outcome would be inaccurate. You can refer to huggingface/peft#1474 where we discuss this.
code: result_dora = (mag_norm_scale - 1) * (F.linear(x, transpose(weight, self.fan_in_fan_out)) ) + mag_norm_scale * lora_B(lora_A(x)) * scaling
Question: what is the effect of (mag_norm_scale - 1) and mag_norm_scale ? And, result_dora can't equals the F.linear(x, transpose(weight, self.fan_in_fan_out)) in the Initializing stage due to the parameter "mag_norm_scale - 1"
The text was updated successfully, but these errors were encountered: