Update modelling_hibial.py

modelling_hibial.py

@@ -147,7 +147,8 @@ class TransformerEncoderLayer(Module):
 
         self.self_attn = MultiheadAttention(config.hidden_size,
                                             config.num_attention_heads,
-                                            dropout=config.attention_probs_dropout_prob)
+                                            dropout=config.attention_probs_dropout_prob,
+                                            dtype = torch.float32)
         # Implementation of Feedforward model
         self.linear1 = Linear(config.hidden_size, config.intermediate_size)
         self.dropout = Dropout(config.hidden_dropout_prob)