ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update llama.py

Browse source
This commit is contained in:
Daniel Han-Chen 2024-02-04 02:02:51 +11:00
commit fcb884643b
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
unsloth/models/llama.py
View file

@ -281,7 +281,7 @@ def LlamaAttention_fast_forward(
past_key_value = (K, V) if use_cache else None
# Attention module
if (not HAS_FLASH_ATTENTION):
if (not HAS_FLASH_ATTENTION and attention_mask is None):
# Xformers memory efficient attention
# Also has Flash Attention v2 dispatching
Q = Q.transpose(1, 2)
@ -303,7 +303,7 @@ def LlamaAttention_fast_forward(
A = xformers_attention(Q, K, V, attn_bias = causal_mask)
A = A.view(bsz, q_len, n_heads, head_dim)
elif HAS_FLASH_ATTENTION:
elif HAS_FLASH_ATTENTION and attention_mask is None:
Q = Q.transpose(1, 2)
K = K.transpose(1, 2)
V = V.transpose(1, 2)