Jasper - fine tuning on pretrained model, with different language/different label list

[jereb321]

When i add some additional labels, into the jasper10x5dr_speedp-online_speca.yaml file and run train.py, i get the following error:

size mismatch for decoder.layers.0.weight: copying a param with shape torch.Size([29, 1024, 1]) from checkpoint, the shape in current model is torch.Size([32, 1024, 1]). size mismatch for decoder.layers.0.bias: copying a param with shape torch.Size([29]) from checkpoint, the shape in current model is torch.Size([32]).

Is there any way to fine tune from the english ASR model, with a different dataset/different label list? I know that there was an --warm_start option for Text to speech Tacotron2 for this specific problem, is here in Jasper maybe some similar parameter to use?

Best regards jereb321

[matko123]

Did you find any solution? Having the same issue/question.