Issues about reproduing results in paper

[TonyStark042]

Hi, thanks for your contribution about CALM repo.
Now i am trying to reproduce the results but the eval/brierLM only reaches 0.04 at around 40k training steps, which mismatches the results reported in the paper.
I am using the deafalt config in train/train_energy.sh with A100*8 and have downloaded the pre-trained AE, no extra changes were made.
So may i ask are there any issues in the code or i misundertood the usage?

[fedemelis]

I had a similar issue while evaluating the provided checkpoints.
The discrepancy was due to the fact that the BrierLM score is reported in [0,1] and needs to be multiplied by 100 to match the expected scale.

[goodluckcwl]

Hi, thanks for your contribution about CALM repo. Now i am trying to reproduce the results but the eval/brierLM only reaches 0.04 at around 40k training steps, which mismatches the results reported in the paper. I am using the deafalt config in train/train_energy.sh with A100*8 and have downloaded the pre-trained AE, no extra changes were made. So may i ask are there any issues in the code or i misundertood the usage?

The same