Update modeling_internlm2.py

modeling_internlm2.py

@@ -460,7 +460,7 @@ class InternLM2Attention(nn.Module):
         
         if self.headcut:
             '''
-            # compute headmask based on attn weights ratio.
+            # compute attn weights ratio for headcut mask generation.
             if attn_weights.shape[2]>1:
                 res_v = torch.sum( attn_weights[0,:,:,self.offset:self.offset+image_token_num],dim=[1,2])
                 res_t = torch.sum( attn_weights[0,:,:,self.offset+image_token_num:],dim=[1,2])