Improvement of huge model transformation (implementation of Out of Memory workaround)

[PINTO0309]

Issue Type

Others

OS

Linux

onnx2tf version number

1.25.7

onnx version number

1.16.1

onnxruntime version number

1.18.1

onnxsim (onnx_simplifier) version number

0.4.33

tensorflow version number

2.17.0

Download URL for ONNX

https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/onnx/unet

Parameter Replacement JSON

NA

Description

OOM often occurs when converting models with very large model structures due to the large number of Numpy Arrays deployed in RAM during automatic correction for accuracy errors.

• Ref
  https://zenn.dev/kazuhito/articles/800e1176270c43

  

  [TODO] Switch to a mode that performs inference in slow delay shape estimation mode during very large size transformations such as diffusion models #430

  https://github.com/Kazuhito00/BiRefNet-ONNX-Sample/releases/download/v0.0.1/birefnet_1024x1024.onnx

  onnxsim birefnet_1024x1024.onnx birefnet_1024x1024.onnx
  onnxsim birefnet_1024x1024.onnx birefnet_1024x1024.onnx
  
  sde4onnx \
  --input_onnx_file_path birefnet_1024x1024.onnx \
  --output_onnx_file_path birefnet_1024x1024.onnx
  
  onnx2tf \
  -i birefnet_1024x1024.onnx \
  -nuo