Enable ONNX export decoder only without need to merge (#1257)
* ONNX export decoder model refactorization
* fix IO bindings
* format
* enable mpt support
* format
* add trust remote code
* fix test
* format
* fix quantization
* add test
* format
* fix optimization
* fix compatibility with legacy models
* fix style
* add export to main_export
* add legacy to ONNX export
* patch model to fix causal lm generation
* add no post process
* remove bloom caching
* fix dynamic axis for position ids
* fix external data
* add model patcher
* format
* fix bart model patcher
* fix model patcher for opt models
* fix format
* add test
* format
* fix ort docker
* add test
* fix bart model patcher
* raise when unsupported model
* add cached file
* add position warning
* fixes
* enable post process after export to remove tied weights
* comment
* remove test
* fix test
* modify model
* remove deprecated use_merged in test
* Add mistral model patcher
* add slow test
* add workflow
