Open ZPLai opened 1 year ago
Hi @ZPLai, yes, you could use neural-fortran for that.
Specifically about loading models pre-trained in Python, you could do that if the models were trained in Keras and saved in the HDF5 format. We may support PyTorch and other Keras formats in the future if there's demand and help to implement it.
See examples of running dense and convolutional networks loaded from pre-trained Keras models.
Hi @ZPLai, yes, you could use neural-fortran for that.
Specifically about loading models pre-trained in Python, you could do that if the models were trained in Keras and saved in the HDF5 format. We may support PyTorch and other Keras formats in the future if there's demand and help to implement it.
See examples of running dense and convolutional networks loaded from pre-trained Keras models.
that‘s fine. I'll try it latter. Thanks
Dear developers, thanks for intersting work. I am considering to use deep learning model into FEM modeling by using ABAQUS. I want to use a deep learning model (that prior trained by other language, such as python) in the subrountine written by Fortran. These subrountines generally were used to model complicated material propeties or complicated loading. And the trained deep learning model may assist for these tasks.
In previous, I thought I may extract the trained model's parameters into these rountine, and then, build the simple fully-connected ANN in fortran. However, if the model is large, such as complicated CNN, the coding could be time-comsuining.
So I wonder if the tool you developed can help this work?