Compress the model generated by dpgen

[jic198]

Hello. I noticed that when I used the model generated by dpgen to run lammps, its speed was much slower than the one generated by deepmd-kit. I typically compressed the model to speed up the MD calculations. Is there a way to compress the model generated by dpgen? I tried dp compress -i graph.pb -o graph-compress.pb but got a following error:
AssertionError: Model compression error: descriptor resnet_dt must be false! Could you help me out? Thank you!

[chunweizhu]

Maybe, you need to change the parameter resnet_dt: false in the param.json file and train the model once again, i.e. without using the residual network.