(AR-Diffusion) RuntimeError: Error(s) in loading state_dict for CrossAttention_Diffusion_LM about prophetnet HOT 3 CLOSED

EMA is a commonly used training method (https://www.fidelity.com/learning-center/trading-investing/technical-analysis/technical-indicator-guide/ema)

If you want to use ema during inference, you also need to add this parameter during training. This code will save two models, one without ema and one with ema.

from prophetnet.

According to your error message, I think it is not caused by ema, but because you added time_trans during training, that is (https://github.com/microsoft/ProphetNet/blob/e2c6657309537b94818f5ddbb2a2c5b5559257bf/AR-diffusion/model_utils/ CrossAttention.py#L31), the corresponding parameter is time_att=True.
I want to set time_att=False to run the code.
In addition, if you want to add time_att=True during inference, you should also add this parameter during training.

from prophetnet.

I want to set time_att=False to run the code.

Perfect, setting time_att=False removes the error during inference. I realised that I didn't use time_att=True during training.

from prophetnet.