Skip to content

⚡️ Speed up function eager_attention_forward by 9% #140

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

⚡️ Speed up function eager_attention_forward by 9% #140

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
8 changes: 5 additions & 3 deletions src/transformers/models/vit_mae/modeling_vit_mae.py
100755 → 100644
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -341,14 +341,16 @@ def eager_attention_forward(
scaling = query.size(-1) ** -0.5

# Take the dot product between "query" and "key" to get the raw attention scores.
attn_weights = torch.matmul(query, key.transpose(2, 3)) * scaling
attn_weights = torch.matmul(query, key.transpose(2, 3))
attn_weights.mul_(scaling)

if attention_mask is not None:
attention_mask = attention_mask[:, :, :, : key.shape[-2]]
attn_weights = attn_weights + attention_mask
attn_weights.add_(attention_mask)

attn_weights = nn.functional.softmax(attn_weights, dim=-1)
attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
if dropout > 0.0:
attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)

attn_output = torch.matmul(attn_weights, value)
attn_output = attn_output.transpose(1, 2).contiguous()
Expand Down