update get_parameter_dtype

[yiyixuxu]

small optimization on dtype method for our ModelMixin
see context in this PR #9520 (comment)

main: Getting unet.dtype took 0.004666426 seconds
PR branch: Getting unet.dtype took 0.000084107 seconds

from diffusers import UNetMotionModel, UNet2DConditionModel, MotionAdapter
import torch

dtype = torch.float32

adapter = MotionAdapter.from_pretrained("guoyww/animatediff-motion-adapter-v1-5-2", torch_dtype=dtype)
# load SD 1.5 based finetuned model
model_id = "SG161222/Realistic_Vision_V5.1_noVAE"
unet = UNet2DConditionModel.from_pretrained(model_id, subfolder="unet", torch_dtype=dtype)

# wrap model with motion adapter
unet = UNetMotionModel.from_unet2d(unet, adapter)
unet.to("cuda")

import time
from contextlib import contextmanager

@contextmanager
def timer(name):
    start_time = time.perf_counter()
    yield
    end_time = time.perf_counter()
    print(f"{name} took {end_time - start_time:.9f} seconds")

with timer("Getting unet.dtype"):
    print(unet.dtype)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[SunMarc]

Nice optimization! LGTM!

[xiang9156]

Thank you guys for lightning action!