dreambooth: fix huggingface#1566: maintain fp32 wrapper when saving a checkpoint to avoid crash when running fp16 (huggingface#1618)

* dreambooth: fix huggingface#1566: maintain fp32 wrapper when saving a checkpoint to avoid crash when running fp16

* dreambooth: guard against passing keep_fp32_wrapper arg to older versions of accelerate. part of fix for huggingface#1566

* Apply suggestions from code review

Co-authored-by: Pedro Cuenca <[email protected]>

* Update examples/dreambooth/train_dreambooth.py

Co-authored-by: Patrick von Platen <[email protected]>
Co-authored-by: Pedro Cuenca <[email protected]>

Author: 3 people

examples/dreambooth/train_dreambooth.py

@@ -1,5 +1,6 @@
 import argparse
 import hashlib
+import inspect
 import itertools
 import math
 import os
@@ -690,10 +691,19 @@ def main(args):
 
                 if global_step % args.save_steps == 0:
                     if accelerator.is_main_process:
+                        # When 'keep_fp32_wrapper' is `False` (the default), then the models are
+                        # unwrapped and the mixed precision hooks are removed, so training crashes
+                        # when the unwrapped models are used for further training.
+                        # This is only supported in newer versions of `accelerate`.
+                        # TODO(Pedro, Suraj): Remove `accepts_keep_fp32_wrapper` when forcing newer accelerate versions
+                        accepts_keep_fp32_wrapper = "keep_fp32_wrapper" in set(
+                            inspect.signature(accelerator.unwrap_model).parameters.keys()
+                        )
+                        extra_args = {"keep_fp32_wrapper": True} if accepts_keep_fp32_wrapper else {}
                         pipeline = DiffusionPipeline.from_pretrained(
                             args.pretrained_model_name_or_path,
-                            unet=accelerator.unwrap_model(unet),
-                            text_encoder=accelerator.unwrap_model(text_encoder),
+                            unet=accelerator.unwrap_model(unet, **extra_args),
+                            text_encoder=accelerator.unwrap_model(text_encoder, **extra_args),
                             revision=args.revision,
                         )
                         save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")