Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Resolve issues between kv_cache and flash attention. #1178

Merged
merged 1 commit into from
Mar 8, 2024
Merged

Resolve issues between kv_cache and flash attention. #1178

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion megatron/model/transformer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -626,6 +626,7 @@ def flash_attention(self, query_layer, key_layer, value_layer):
q_shape = query_layer.shape
k_shape = key_layer.shape
v_shape = value_layer.shape
is_causal = max_seqlen_q == max_seqlen_k
output = self.flash_varlen_qkv_fn(
query_layer.reshape(
(q_shape[0] * q_shape[1], q_shape[2], q_shape[3])
Expand All @@ -641,7 +642,7 @@ def flash_attention(self, query_layer, key_layer, value_layer):
max_seqlen_q,
max_seqlen_k,
softmax_scale=None,
causal=True,
causal=is_causal,
**extra_kwargs,
)
output = output.reshape(q_shape)
Expand Down
Loading