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.
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
dreambooth if docs - stage II, more info #3628
dreambooth if docs - stage II, more info #3628
Conversation
The documentation is not available anymore as the PR was closed or merged. |
a0bbc82
to
83b3d83
Compare
83b3d83
to
ab39f6b
Compare
pipe.load_lora_weights("<lora weights path>") | ||
|
||
# Update scheduler config to fixed variance schedule | ||
pipe.scheduler = pipe.scheduler.__class__.from_config(pipe.scheduler.config, variance_type="fixed_small") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
🧠
For common and/or not-visually complex object concepts, you can get away with not-finetuning the upscaler. Just be sure to adjust the prompt passed to the | ||
upscaler to remove the new token from the instance prompt. I.e. if your stage I prompt is "a sks dog", use "a dog" for your stage II prompt. | ||
|
||
For finegrained detail like faces that aren't present in the original training set, we find that full finetuning of the stage II upscaler is better than |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice info!
I would enumerate the tips and tricks to improve readability.
Also, if we can include how the upscaler is supposed to be fine-tuning or even just leave a link that shows it, that would be info-complete.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmm, what should we link to? The stage II fine tuning section is just a bit further on in the page
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Changes LGTM! I left some minor nits.
If possible, I would spend one/two sentences to let the users know what does stage II entail. If it's exactly same as stage I, then nope, otherwise, I think including it makes sense.
pipe.load_lora_weights("<lora weights path>") | ||
|
||
# Update scheduler config to fixed variance schedule | ||
pipe.scheduler = pipe.scheduler.__class__.from_config(pipe.scheduler.config, variance_type="fixed_small") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
pipe.scheduler = pipe.scheduler.__class__.from_config(pipe.scheduler.config, variance_type="fixed_small") | |
pipe.scheduler.register_to_config(variance_type="fixed_small") |
should suffice no?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I wouldn't recommend mutating existing constructed schedulers since depending on the argument they sometimes change other instance variables. Since they're cheap to construct, I think it's good to always demonstrate freshly constructing a new instance
export INSTANCE_DIR="dog" | ||
export OUTPUT_DIR="dreambooth_dog_upscale" | ||
export VALIDATION_IMAGES="image_1.png image_2.png image_3.png image_4.png" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same here where do the dirs and images come from? Let's make it easy for the user
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The dog
directory is downloaded earlier in the README. The OUTPUT_DIR is created automatically by the training script. I think these are consistent with the other snippets
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* dreambooth if docs - stage II, more info * Update docs/source/en/training/dreambooth.mdx Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> * Update docs/source/en/training/dreambooth.mdx Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> * Update docs/source/en/training/dreambooth.mdx Co-authored-by: Sayak Paul <spsayakpaul@gmail.com> * download instructions for downsized images * update source README to match docs --------- Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com> Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
More fleshed out docs we can link to in model cards and IF readme