Skip to content

Fix attention layers map for SD-2-1-Base #971

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 2 commits into
base: main
Choose a base branch
from
Open

Fix attention layers map for SD-2-1-Base #971

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1fbc936
Fix attention layers map for SD-2-1-Base
pranav4501 Sep 1, 2024
282304a
fix mapping for sdxl turbo
pranav4501 Sep 4, 2024
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
11 changes: 11 additions & 0 deletions stable_diffusion/stable_diffusion/model_io.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -140,6 +140,17 @@ def map_vae_weights(key, value):
if "to_v" in key:
key = key.replace("to_v", "value_proj")


# Map attention layers in SD-2-1-base:VAE
if "key" in key:
key = key.replace("key", "to_k")
if "proj_attn" in key:
key = key.replace("proj_attn", "out_proj")
if "query" in key:
key = key.replace("query", "query_proj")
if "value" in key:
key = key.replace("value", "value_proj")
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

These don't look correct given that they will result in value_proj_proj and query_proj_proj and to_k. Am I missing sth?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, I missed that. I didn't check for sdxl turbo. It should be fine now if we first check for query and then for to_q.
Also, why are we not using load_weights here? Doesn't it help with such mapping issues?


# Map the mid block
if "mid_block.resnets.0" in key:
key = key.replace("mid_block.resnets.0", "mid_blocks.0")
Expand Down