Update modeling_opt.py

modeling_opt.py

@@ -227,7 +227,6 @@ class OPTAttention(nn.Module):
             attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
 
         # upcast to fp32 if the weights are in fp16. Please see https://github.com/huggingface/transformers/pull/17437
-        breakpoint()
         if attn_weights.dtype == torch.float16:
             attn_weights = self.softmax(attn_weights.float()).to(torch.float16)
         else: