Add Stable Diffusion ONNX export (#570)
* Add stable diffusion UNET model ONNX export
* Add possibility to give a dict config when instantiating a NormalizedConfig
* Add clip and vae onnx export
* Fix style
* Add stable diffusion text encoder and vae model onnx configuration
* Add dummy inputs generation
* Fix style
* Add stable diffusion export ONNX CLI
* Remove unsupported dummy input name
* Fix style
* Fix default atol
* Refactorization
* Add onnxruntime output names for comparison
* Add test
* Add import utils
* Add model decomposition outside export
* Clean import utilities
* Fix style
* Add missing import
* Apply suggestions
* Apply suggestions
* Infer stable-diffusion when auto
* Apply suggestions
* Save stable diffusion model tokenizer
* Fix test
* Improve docstring
* Change name parameter for clarity
* Fix style
* Apply suggestions
* Fix style
* Fix clip text model type
* Take model type from config when possible
* Update docstring
* Apply suggestion
* Fix test temporarily
* Add supported model by CLI
* Remove unused import
* Fix ONNX export for layoutlmv3 architecture
* Make atol and opset optional
* Update optimum/exporters/onnx/convert.py
Co-authored-by: fxmarty <9808326+fxmarty@users.noreply.github.com>
* Update optimum/exporters/onnx/convert.py
Co-authored-by: fxmarty <9808326+fxmarty@users.noreply.github.com>
* Update optimum/exporters/onnx/convert.py
Co-authored-by: fxmarty <9808326+fxmarty@users.noreply.github.com>
* Update optimum/exporters/onnx/convert.py
Co-authored-by: fxmarty <9808326+fxmarty@users.noreply.github.com>
* Add stable diffusion model ONNX export in documentation
Co-authored-by: fxmarty <9808326+fxmarty@users.noreply.github.com>
