Improve LCM(-LoRA) Distillation Scripts #6420
Merged
Conversation
This file contains hidden or 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
|
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
reviewed
Jan 2, 2024
Member
sayakpaul
left a comment
sayakpaul
left a comment
There was a problem hiding this comment.
LGTM, thanks!
- Let's also support interpolation mode in the non-WDS script?
- Might make sense to add
resolve_interpolation_mode()intraining_utils.py, as I imagine it being used in multiple scripts.
Contributor
|
@patil-suraj can you check here? |
…d make interpolation type configurable in non-WDS script.
patil-suraj
approved these changes
Jan 4, 2024
examples/consistency_distillation/train_lcm_distill_lora_sd_wds.py
Outdated
Show resolved
Hide resolved
examples/consistency_distillation/train_lcm_distill_lora_sdxl.py
Outdated
Show resolved
Hide resolved
examples/consistency_distillation/train_lcm_distill_lora_sdxl_wds.py
Outdated
Show resolved
Hide resolved
Member
|
Thanks a lot, @dg845! The scripts are in a really good spot now! |
dg845
added a commit
to dg845/diffusers
that referenced
this pull request
Jan 7, 2024
yiyixuxu
reviewed
Jan 31, 2024
|
|
||
| import numpy as np | ||
| import torch | ||
| from torchvision import transforms |
Collaborator
There was a problem hiding this comment.
we can't have torchvision as a required dependency
AmericanPresidentJimmyCarter
pushed a commit
to AmericanPresidentJimmyCarter/diffusers
that referenced
this pull request
Apr 26, 2024
* Make WDS pipeline interpolation type configurable. * Make the VAE encoding batch size configurable. * Make lora_alpha and lora_dropout configurable for LCM LoRA scripts. * Generalize scalings_for_boundary_conditions function and make the timestep scaling configurable. * Make LoRA target modules configurable for LCM-LoRA scripts. * Move resolve_interpolation_mode to src/diffusers/training_utils.py and make interpolation type configurable in non-WDS script. * apply suggestions from review
|
Wether TCD distilltion training is supported by Diffusers |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
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?
This PR makes further improvements to the LCM-(LoRA) distillation scripts.
Changelist:
--interpolation_typeargument.lora_alpha(which controls LoRA scaling) andlora_dropoutparameters configurable via the--lora_alphaand--lora_dropoutarguments respectively.--lora_target_modulesargument.--vae_encode_batch_sizeargument.scalings_for_boundary_conditionsfunction to use thetimestep_scalingargument (instead of it being hardcoded to10.0) and make the scaling factor configurable via--timestep_scaling_factor.Follow-up to #5778.
Before submitting
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
@patrickvonplaten
@sayakpaul
@patil-suraj