Size mismatch error with Pytorch's TorchEntityRecognizer on evaluation

[felipefoschiera]

I am having some trouble evaluating a transformer model (roberta-base) trained with the torch_ner pipeline.
I was able to run both train+evaluation commands with spaCy's ner + transformer, as well as torch_ner + vectors. But for torch_ner + transformer, I was only able to train, and I get the following error when trying to evaluate the model:

RuntimeError: Error(s) in loading state_dict for TorchEntityRecognizer:
        size mismatch for model.input_layer.weight: copying a param with shape torch.Size([64, 768]) from checkpoint, the shape in current model is torch.Size([64, 1]).

The command executed was
python -m spacy evaluate training/training_pytorch_trf/model-best corpus/test.spacy --output training/training_pytorch_trf/metrics.json --code scripts/custom_functions.py --gpu-id 0

The config.cfg file I am using is attached to this discussion.
config_pytorch_trf_txt.txt

[svlandeg]

Hi! Thanks for the report and sorry for the late follow-up. We're working on a fix: explosion/projects#80