tiny fix

Former-commit-id: 7451b2ae7e58d0f1857f01a037672a8c53b1bd0d
2023-09-28 01:02:11 +08:00
parent aecbb43096
commit 22c6c27f78
2 changed files with 2 additions and 2 deletions
--- a/src/llmtuner/extras/patches/llama_patch.py
+++ b/src/llmtuner/extras/patches/llama_patch.py
@@ -173,7 +173,7 @@ class LlamaFlashAttention2(LlamaAttention):
                state = state.reshape(bsz * num_group, group_size, self.num_heads, self.head_dim)

        if attention_mask is not None:
-            logger.warning_once("Padded sequences are less efficient.")
+            logger.warning_once("Padded sequences are less efficient in FlashAttention.")
            batch_size = query_states.shape[0]
            # -q_len: assumes left padding
            unpadded_q, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(query_states, attention_mask[:, -q_len:])