[LoRA] Remove the use of depcrecated loRA functionalities such as LoRAAttnProcessor
#6369
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Merged
sayakpaul
commented
Dec 28, 2023
| @@ -311,71 +355,12 @@ def test_stable_diffusion_xformers_attn_processors(self): | |||
| image = sd_pipe(**inputs).images | |||
| assert image.shape == (1, 64, 64, 3) | |||
|
|
|||
| @unittest.skipIf(not torch.cuda.is_available(), reason="xformers requires cuda") | |||
There was a problem hiding this comment.
Don't think we need to keep this test because it uses attention processors. Objections?
sayakpaul
commented
Dec 28, 2023
| assert image.shape == (1, 64, 64, 3) | ||
|
|
||
| @unittest.skipIf(not torch.cuda.is_available() or not is_xformers_available(), reason="xformers requires cuda") | ||
| def test_stable_diffusion_set_xformers_attn_processors(self): |
sayakpaul
commented
Dec 28, 2023
| lora_attn_procs = create_lora_layers(sd_pipe.unet) | ||
| sd_pipe.unet.set_attn_processor(lora_attn_procs) | ||
| sd_pipe = sd_pipe.to(torch_device) | ||
| sd_pipe.unet.load_attn_procs(lora_components["unet_lora_params"]) |
There was a problem hiding this comment.
It still works on regular state dicts as expected.
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
sayakpaul
commented
Dec 28, 2023
| @@ -1480,120 +1502,6 @@ def test_lora_processors(self): | |||
| # sample 2 and sample 3 should be different | |||
| assert (sample2 - sample3).abs().max() > 1e-4 | |||
|
|
|||
| def test_lora_save_load(self): | |||
There was a problem hiding this comment.
Same. No need to test for attention processors I think.
sayakpaul
commented
Dec 28, 2023
| @@ -1087,7 +1129,7 @@ def test_lora_fuse_nan(self): | |||
| # corrupt one LoRA weight with `inf` values | |||
| with torch.no_grad(): | |||
| sd_pipe.unet.mid_block.attentions[0].transformer_blocks[0].attn1.to_q.lora_layer.down.weight += float( | |||
| "inf" | |||
| "NaN" | |||
There was a problem hiding this comment.
The reason why we do this is the following:
import torch
a = torch.tensor([float("inf")])
print(torch.isnan(a).any().item())Prints False, whereas replacing the "inf" with "NaN" prints True.
There was a problem hiding this comment.
Ok for me, but even adding an "inf" value should yield NaNs for the forward pass
patrickvonplaten
approved these changes
Jan 2, 2024
AmericanPresidentJimmyCarter
pushed a commit
to AmericanPresidentJimmyCarter/diffusers
that referenced
this pull request
Apr 26, 2024
…RAAttnProcessor` (huggingface#6369) * start deprecating loraattn. * fix * wrap into unet_lora_state_dict * utilize text_encoder_lora_params * utilize text_encoder_attn_modules * debug * debug * remove print * don't use text encoder for test_stable_diffusion_lora * load the procs. * set_default_attn_processor * fix: set_default_attn_processor call. * fix: lora_components[unet_lora_params] * checking for 3d. * 3d. * more fixes. * debug * debug * debug * debug * more debug * more debug * more debug * more debug * more debug * more debug * hack. * remove comments and prep for a PR. * appropriate set_lora_weights() * fix * fix: test_unload_lora_sd * fix: test_unload_lora_sd * use dfault attebtion processors. * debu * debug nan * debug nan * debug nan * use NaN instead of inf * remove comments. * fix: test_text_encoder_lora_state_dict_unchanged * attention processor default * default attention processors. * default * style
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
More concrete problem that this PR solves: https://github.com/huggingface/diffusers/actions/runs/7339182559/job/19982940578.
In summary, we remove: