Fix ORT CI (#1875)
* fix quantizer
* change diffusion test
* install cpu torch
* fix
* fix only for qdq quantizer
* fix past kv in old model
* warn
* assert text equal
* Update optimum/onnxruntime/modeling_decoder.py
Co-authored-by: Michael Benayoun <mickbenayoun@gmail.com>
* use action to free disk
* change input text
* keep large packages
* update python
* test with original gpt2 tokenizer
* test
* run everything
* Update tests/onnxruntime/test_modeling.py
Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>
* make old onnx model inference tests run in series as they modify the underlying model
* my bad
* compare optimum ort sd with diffusers onnx sd because they're the only ones using np random states
* seperate ort test subsets to propagate each subsets' return code
---------
Co-authored-by: Ella Charlaix <ella@huggingface.co>
Co-authored-by: Michael Benayoun <mickbenayoun@gmail.com>
Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>
