locuslab/massive-activations

Issues

• the standard deviation of the activation

  #7 opened 3 months ago by Cooperx521
  3

• Thoughts about ViT register and anchor token

  #6 opened 3 months ago by Cooperx521
  3

• Training only on 2B tokens (openwebtext)

  #5 opened 6 months ago by Nandan91
  3

• what if set all layers' massive activations to their mean value?

  #4 opened 7 months ago by yyfcc17
  1

• How to get the mean value of massive activation

  #3 opened 7 months ago by pengyao96
  1

• Which layer's activation is used?

  #1 opened 7 months ago by iyupan
  1