katuni4ka commited on
Commit
9db1a41
·
verified ·
1 Parent(s): d6cb7c8

Update modeling_baichuan.py

Browse files
Files changed (1) hide show
  1. modeling_baichuan.py +2 -2
modeling_baichuan.py CHANGED
@@ -230,8 +230,8 @@ class Attention(nn.Module):
230
  query_states, key_states, value_states, attn_bias=xops.LowerTriangularMask()
231
  )
232
  else:
233
- with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=True, enable_mem_efficient=True):
234
- attn_output = F.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask = attention_mask)
235
  attn_output = attn_output.transpose(1, 2)
236
  attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
237
  attn_output = self.o_proj(attn_output)
 
230
  query_states, key_states, value_states, attn_bias=xops.LowerTriangularMask()
231
  )
232
  else:
233
+ #with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=True, enable_mem_efficient=True):
234
+ attn_output = F.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask = attention_mask)
235
  attn_output = attn_output.transpose(1, 2)
236
  attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
237
  attn_output = self.o_proj(attn_output)