Anyang Peng

added in this PR https://github.com/deepmodeling/deepmd-kit/pull/3266/

> > repeatedly each time it is fine-tuned > > Indeed, I don't understand why it needs to change the type map each time it is fine-tuned... For the LinearModel,...

> > > > repeatedly each time it is fine-tuned > > > > > > > > > > > > > > Indeed, I don't understand why it...

@wanghan-iapcm feels like we can leave this unchanged based on our discussion. 1. during training, user knows what `type_map` to use, thus, three models should share the same `type_map`. 2....

I may close this PR if ndarry is the expected output type.

> Could you please explain the reason of using `Tuple[np.ndarray]` instead of `np.ndarray` as returned type of `DeepTensor`? It seems the other DeepModels all return a tuple object, I thought...

@shyuep Please check whether the formula and units used in calculating Clarke's thermal conductivity are correct. According to GPT, a Reduced Planck constant is missing.

following up on this

@njzjz Can you add the JAX backend and see if it's consistent with either TF or PT?

> > [@njzjz](https://github.com/njzjz) Can you add the JAX backend and see if it's consistent with either TF or PT? > > The JAX backend reuses codes from the DP backend....