Skip to content
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.

Sign up for GitHub

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

Fix custom ops loading in diffusers #1655

Merged
merged 3 commits into from
Jan 17, 2025
Merged

Fix custom ops loading in diffusers #1655

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e80b854
Fix custom ops loading in diffusers
dsocek Dec 20, 2024
e8dad7d
Fix gaudi config type handling
dsocek Jan 10, 2025
bd91d31
Merge remote-tracking branch 'optimum-habana/main' into fix-diffusers…
regisss Jan 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 0 additions & 4 deletions examples/stable-diffusion/text_to_image_generation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -440,10 +440,6 @@ def main():

kwargs_call["quant_mode"] = args.quant_mode

if args.quant_mode != "disable":
# Import htcore here to support model quantization
import habana_frameworks.torch.core as htcore # noqa: F401

# Instantiate a Stable Diffusion pipeline class
if sdxl:
# SDXL pipelines
Expand Down
13 changes: 11 additions & 2 deletions optimum/habana/diffusers/pipelines/pipeline_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -165,8 +165,6 @@ def __init__(
"`torch_dtype=torch.bfloat16` was given. Disabling mixed precision and continuing in bf16 only."
)
self.gaudi_config.use_torch_autocast = False
else:
self.gaudi_config.declare_autocast_bf16_fp32_ops()

dsocek marked this conversation as resolved.
Show resolved Hide resolved
# Workaround for Synapse 1.11 for full bf16 and Torch Autocast
if bf16_full_eval or self.gaudi_config.use_torch_autocast:
Expand Down Expand Up @@ -370,6 +368,17 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
bf16_full_eval = kwargs.get("torch_dtype", None) == torch.bfloat16
kwargs["bf16_full_eval"] = bf16_full_eval

# Need to load custom ops lists before instantiating htcore
if kwargs.get("gaudi_config", None) is not None:
if isinstance(kwargs["gaudi_config"], str):
gaudi_config = GaudiConfig.from_pretrained(kwargs["gaudi_config"])
else:
gaudi_config = kwargs["gaudi_config"]
gaudi_config.declare_autocast_bf16_fp32_ops()
kwargs["gaudi_config"] = gaudi_config

# Import htcore here to support model quantization
import habana_frameworks.torch.core as htcore # noqa: F401
return super().from_pretrained(
pretrained_model_name_or_path,
**kwargs,
Expand Down