liusx commited on
Commit
59f0181
·
verified ·
1 Parent(s): a6ffa79

fix infer bug while use_flash_attention=False

Browse files
Files changed (1) hide show
  1. modeling_telechat.py +1 -1
modeling_telechat.py CHANGED
@@ -257,7 +257,7 @@ class TELECHATAttention(nn.Module):
257
  self.pruned_heads = set()
258
 
259
  self.use_flash_attn = False
260
-
261
 
262
 
263
  def set_max_positions(self, max_positions, device='cuda'):
 
257
  self.pruned_heads = set()
258
 
259
  self.use_flash_attn = False
260
+ self.is_cross_attention = False
261
 
262
 
263
  def set_max_positions(self, max_positions, device='cuda'):