fix llama fa bug (PaddlePaddle#8237)

DesmonDay · Apr 9, 2024 · 79cb8b6 · 79cb8b6
1 parent c1aad02
commit 79cb8b6
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/paddlenlp/transformers/llama/modeling.py b/paddlenlp/transformers/llama/modeling.py
@@ -1550,7 +1550,7 @@ def forward(
             attention_mask, (batch_size, seq_length), cache_length, inputs_embeds.dtype
         )  # [bs, 1, seq_len, seq_len]
         if self.config.use_flash_attention:
-            if get_env_device != "npu":
+            if get_env_device() != "npu":
                 is_casual = is_casual_mask(attention_mask)
                 if is_casual and alibi is None:
                     attention_mask = None