This pr should allow for some more options such as exporting all models in one go, the desired precision ( fp16 / fp32, TensorRT 10 sometimes throws a fit with the exports' precision ), desired opset if wanted ( 19 is the max with torch 2.2 but the nightly builds allow for opset 20 as well ) and also allow for model optimizations through ONNXSlim.
Hi,
This pr should allow for some more options such as exporting all models in one go, the desired precision ( fp16 / fp32, TensorRT 10 sometimes throws a fit with the exports' precision ), desired opset if wanted ( 19 is the max with torch 2.2 but the nightly builds allow for opset 20 as well ) and also allow for model optimizations through ONNXSlim.