423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-15 11:20:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

tiny fix

Browse Source
This commit is contained in:
hiyouga
2023-09-28 01:02:11 +08:00
parent c902236397
commit d2ebd225db
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llmtuner/extras/patches/llama_patch.py
View File

@@ -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:])