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

Few fixs of Moat #150

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
+12 −4
Open

Few fixs of Moat #150

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f7b9698
Typo fixed: _make_wondows -> _make_windows
edwardyehuang Nov 11, 2022
63d2a23
Fixed global window implementation
edwardyehuang Nov 11, 2022
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
16 changes: 12 additions & 4 deletions model/layers/moat_blocks.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -460,7 +460,7 @@ def build(self, input_shape: list[int]) -> None:
kernel_initializer=self._config.kernel_initializer,
bias_initializer=self._config.bias_initializer)

def _make_wondows(self, inputs):
def _make_windows(self, inputs):
_, height, width, channels = inputs.get_shape().with_rank(4).as_list()
inputs = tf.reshape(
inputs,
Expand Down Expand Up @@ -530,10 +530,18 @@ def call(self, inputs: tf.Tensor, training: bool) -> tf.Tensor:
attention_shortcut = output
def _func(output):
output = self._attention_norm(output)
_, height, width, _ = output.get_shape().with_rank(4).as_list()
output = self._make_wondows(output)
_, height, width, channels = output.get_shape().with_rank(4).as_list()

if self._config.window_size:
output = self._make_windows(output)

output = self._attention(output)
output = self._remove_windows(output, height, width)

if self._config.window_size:
output = self._remove_windows(output, height, width)
else:
output = tf.reshape(output, [-1, height, width, channels])

return output

func = _func
Expand Down