optimum

16618fc9 - Introduce ORTSessionMixin and enable general io binding (works for diffusers as well) (#2234)

Introduce ORTSessionMixin and enable general io binding (works for diffusers as well) (#2234) * refactor ort session mixin and enable integral and simple diffusers io binding as a result * style * distribute onnxruntime tests * no need to clean disk for fast tests * move diffusion tests to diffusion * fix * test * providers * fix * fix and get rid of io bindign helpers * get rid of the models folder in onnxruntime * style * comments * remove _from_transformers * fix trust remote code * fix local model test (use_cache only should be pased to causal lm) * reuse input buffers as output buffers in diffusion models * fix sess_option * override libarary for sentence transformers feature etraction * fix no cross attention * remove lib and default to transformers * better * flaky decoder * add saving session utils * test * more refactoring attempt (seperating encoder/decode parts deom parent model) * remove print * fix * _infer_onnx_filename as private method * fix model name * fix more model paths * fixes * Update optimum/onnxruntime/base.py Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com> * added main export library guards, and restricted when to force eager * style * review suggestions and added warning when properties are not consistent across * ORTParentMixin * deprecate instantiating ORTModel, ORTModelForCausalLM and ORTModelForConditionalGeneration with positional arguments. * keyword arguments * diffusion * known output buffers * style * typo * id * fix * more typos * Update optimum/onnxruntime/modeling_diffusion.py Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com> * remove task argument from _export * deprecate and fix * fix * style * slim later * misc fixes and extensions in testing * allow passing export arguments to diffusion pipeline (e.g. exports on cuda device, with specific dtype, etc) --------- Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>