Min-SNR Gamma: correct the fix for SNR weighted loss in v-prediction by adding 1 to SNR rather than the resulting loss weights
Co-authored-by: bghira <bghira@users.github.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* fix: #4206
* add: sdxl controlnet training smoketest.
* remove unnecessary token inits.
* add: licensing to model card.
* include SDXL licensing in the model card and make public visibility default
* debugging
* debugging
* disable local file download.
* fix: training test.
* fix: ckpt prefix.
* add: controlnet sdxl.
* modifications to controlnet.
* run styling.
* add: __init__.pys
* incorporate https://github.com/huggingface/diffusers/pull/4019 changes.
* run make fix-copies.
* resize the conditioning images.
* remove autocast.
* run styling.
* disable autocast.
* debugging
* device placement.
* back to autocast.
* remove comment.
* save some memory by reusing the vae and unet in the pipeline.
* apply styling.
* Allow low precision sd xl
* finish
* finish
* changes to accommodate the improved VAE.
* modifications to how we handle vae encoding in the training.
* make style
* make existing controlnet fast tests pass.
* change vae checkpoint cli arg.
* fix: vae pretrained paths.
* fix: steps in get_scheduler().
* debugging.
* debugging./
* fix: weight conversion.
* add: docs.
* add: limited tests./
* add: datasets to the requirements.
* update docstrings and incorporate the usage of watermarking.
* incorporate fix from #4083
* fix watermarking dependency handling.
* run make-fix-copies.
* Empty-Commit
* Update requirements_sdxl.txt
* remove vae upcasting part.
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* run make style
* run make fix-copies.
* disable suppot for multicontrolnet.
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* run make fix-copies.
* dtyle/.
* fix-copies.
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
controlnet training center crop input images to multiple of 8
The pipeline code resizes inputs to multiples of 8.
Not doing this resizing in the training script is causing
the encoded image to have different height/width dimensions
than the encoded conditioning image (which uses a separate
encoder that's part of the controlnet model).
We resize and center crop the inputs to make sure they're the
same size (as well as all other images in the batch). We also
check that the initial resolution is a multiple of 8.
* WIP controlnet training
- bugfix --streaming
- bugfix running report_to!='wandb'
- adds memory profile before validation
* Adds final logging statement.
* Sets train epochs to 11.
Looking at a longer ~16ep run, we see only good validation images
after ~11ep:
https://wandb.ai/andsteing/controlnet_fill50k/runs/3j2hx6n8
* Removes --logging_dir (it's not used).
* Adds --profile flags.
* Updates --output_dir=runs/fill-circle-{timestamp}.
* Compute mean of `train_metrics`.
Previously `train_metrics[-1]` was logged, resulting in very bumpy train
metrics.
* Improves logging a bit.
- adds l2_grads gradient norm logging
- adds steps_per_sec
- sets walltime as x coordinate of train/step
- logs controlnet_params config
* Adds --ccache (doesn't really help though).
* minor fix in controlnet flax example (#2986)
* fix the error when push_to_hub but not log validation
* contronet_from_pt & controlnet_revision
* add intermediate checkpointing to the guide
* Bugfix --profile_steps
* Sets `RACKER_PROJECT_NAME='controlnet_fill50k'`.
* Logs fractional epoch.
* Adds relative `walltime` metric.
* Adds `StepTraceAnnotation` and uses `global_step` insetad of `step`.
* Applied `black`.
* Streamlines commands in README a bit.
* Removes `--ccache`.
This makes only a very small difference (~1 min) with this model size, so removing
the option introduced in cdb3cc.
* Re-ran `black`.
* Update examples/controlnet/README.md
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* Converts spaces to tab.
* Removes repeated args.
* Skips first step (compilation) in profiling
* Updates README with profiling instructions.
* Unifies tabs/spaces in README.
* Re-ran style & quality.
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* ⚙️chore(train_controlnet) fix typo in logger message
* ⚙️chore(models) refactor modules order; make them the same as calling order
When printing the BasicTransformerBlock to stdout, I think it's crucial that the attributes order are shown in proper order. And also previously the "3. Feed Forward" comment was not making sense. It should have been close to self.ff but it's instead next to self.norm3
* correct many tests
* remove bogus file
* make style
* correct more tests
* finish tests
* fix one more
* make style
* make unclip deterministic
* ⚙️chore(models/attention) reorganize comments in BasicTransformerBlock class
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* ensure validation image RGB not RGBA
* ensure validation image RGB not RGBA
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
