How to save intermediate callback variables in inverse analysis?

[wangxi0706]

I am using Pytorch 2.2.2 as the backend. I define the variable callback as follows:

model.compile("L-BFGS", external_trainable_variables=cv)
variable = dde.callbacks.VariableValue(cv, period=1000, filename="cv.txt")
losshistory, train_state = model.train(callbacks=[variable])

However, after 4000 iterations, only the initial and final values are wirrten out. How could I get the ones during iterations?

[lululxvi]

Use another optimizer.