fix: Fix train_dreambooth_lora.py fp16 unscale error after validation - #14507
Open
Ricardo-M-L wants to merge 1 commit into
Open
fix: Fix train_dreambooth_lora.py fp16 unscale error after validation#14507Ricardo-M-L wants to merge 1 commit into
train_dreambooth_lora.py fp16 unscale error after validation#14507Ricardo-M-L wants to merge 1 commit into
Conversation
When `--mixed_precision=fp16` and `--validation_prompt` are both set, training aborts on the first step after the first validation with: ValueError: Attempting to unscale FP16 gradients. Root cause: * The LoRA trainable params are upcast to fp32 once, before training, via `cast_training_params(models, dtype=torch.float32)`. * Validation builds `DiffusionPipeline.from_pretrained(unet=unwrap_model(unet), torch_dtype=weight_dtype, ...)` and hands the pipeline to `log_validation`. * `log_validation` calls `pipeline.to(accelerator.device, dtype=torch_dtype)`, which casts the *shared* `unet` module — including the LoRA adapter weights registered as trainable — back down to fp16. * The next backward then produces fp16 grads, and the grad scaler refuses to unscale them. Re-run `cast_training_params(..., dtype=torch.float32)` immediately after `log_validation` returns (only when `mixed_precision == "fp16"`), mirroring the pre-training upcast. bf16 mixed-precision is unaffected since no grad scaler is in play there. Fixes huggingface#13124
Contributor
|
Hi @Ricardo-M-L, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Clean rebase of fix/dreambooth-lora-fp16-validation-unscale