Dimensions Mismatch

[Benjamin025]

Hello everyone,

I have been trying to run some inferences with the GenCast Mini Model, but during the predictions part I keep getting an error on dimensions mismatch between the checkpoint provided for the mini model and the way my script is processing the variables. Here is the exact error: ValueError: 'grid2mesh_gnn/_networks_builder/encoder_nodes_grid_nodes_mlp//linear_0/w' with retrieved shape (267, 512) does not match shape=[333, 512] dtype=dtype('float32').

This is similar to the issue raised in here:

Looking forward to some engaging discussions.

Thanks.

[Benjamin025]

I meant the same issue was raised in the conversation #131

[alvarosg]

This would usually indicate that the inputs, forcings and targets_template have not been built correctly. Could you verify that the variables, dimension sizes and coordinates of the inputs, forcings and targets_template that you are passing to the model match exactly the sample data provided?

[Benjamin025]

Thanks Alvaro for the reply.

I noticed some inconsistencies in the built in dimensions for the inputs, forcings and target_template, working on it I will update on the progress.