Skip to content

Fix NotImplemented errors, xformers attention shape, and missing text conditioning #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
Mr-Neutr0n wants to merge 1 commit into
base: main
Choose a base branch
from
+9 −3
Open

Fix NotImplemented errors, xformers attention shape, and missing text conditioning #140

Show file tree
Hide file tree
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
4 changes: 3 additions & 1 deletion models/latte.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,7 +70,7 @@ def forward(self, x):
x = (attn @ v).transpose(1, 2).reshape(B, N, C)

else:
raise NotImplemented
raise NotImplementedError

x = self.proj(x)
x = self.proj_drop(x)
Expand Down Expand Up @@ -369,6 +369,8 @@ def forward(self,

if self.extras == 2:
c = timestep_spatial + y_spatial
elif self.extras == 78:
c = timestep_spatial + text_embedding_spatial
else:
c = timestep_spatial
x = self.final_layer(x, c)
Expand Down
8 changes: 6 additions & 2 deletions models/latte_img.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,11 @@ def forward(self, x):
q, k, v = qkv.unbind(0) # make torchscript happy (cannot use tensor as tuple)

if self.attention_mode == 'xformers': # cause loss nan while using with amp
x = xformers.ops.memory_efficient_attention(q, k, v).reshape(B, N, C)
# https://github.com/facebookresearch/xformers/blob/e8bd8f932c2f48e3a3171d06749eecbbf1de420c/xformers/ops/fmha/__init__.py#L135
q_xf = q.transpose(1,2).contiguous()
k_xf = k.transpose(1,2).contiguous()
v_xf = v.transpose(1,2).contiguous()
x = xformers.ops.memory_efficient_attention(q_xf, k_xf, v_xf).reshape(B, N, C)

elif self.attention_mode == 'flash':
# cause loss nan while using with amp
Expand All @@ -73,7 +77,7 @@ def forward(self, x):
x = (attn @ v).transpose(1, 2).reshape(B, N, C)

else:
raise NotImplemented
raise NotImplementedError

x = self.proj(x)
x = self.proj_drop(x)
Expand Down