Skip to content

Optimize causal mask using torch.where #2715

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

Merged
merged 4 commits into from
Apr 7, 2020
Merged

Optimize causal mask using torch.where #2715

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
96efd53
Optimize causal mask using torch.where
Akababa Feb 2, 2020
847a351
Maintain compatiblity with torch 1.0.0 - thanks for PR feedback
Akababa Mar 29, 2020
c2fa05f
Fix typo
Akababa Mar 29, 2020
a54a418
reformat line for CI
Akababa Mar 29, 2020
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
9 changes: 6 additions & 3 deletions src/transformers/modeling_gpt2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,7 +104,10 @@ def __init__(self, nx, n_ctx, config, scale=False):
n_state = nx # in Attention: n_state=768 (nx=n_embd)
# [switch nx => n_state from Block to Attention to keep identical to TF implem]
assert n_state % config.n_head == 0
self.register_buffer("bias", torch.tril(torch.ones(n_ctx, n_ctx)).view(1, 1, n_ctx, n_ctx))
self.register_buffer(
"bias", torch.tril(torch.ones((n_ctx, n_ctx), dtype=torch.uint8)).view(1, 1, n_ctx, n_ctx)
)
self.register_buffer("masked_bias", torch.tensor(-1e4))
self.n_head = config.n_head
self.split_size = n_state
self.scale = scale
Expand Down Expand Up @@ -142,8 +145,8 @@ def _attn(self, q, k, v, attention_mask=None, head_mask=None):
if self.scale:
w = w / math.sqrt(v.size(-1))
nd, ns = w.size(-2), w.size(-1)
b = self.bias[:, :, ns - nd : ns, :ns]
w = w * b - 1e4 * (1 - b)
mask = self.bias[:, :, ns - nd : ns, :ns]
w = torch.where(mask, w, self.masked_bias)

if attention_mask is not None:
# Apply the attention mask
Expand Down