The pre-trained models have effectively become not usable anymore since updates were made to JasperEncoder
or most probable the jasper.py
module.
jasper_encoder = nemo_asr.JasperEncoder(
jasper=jasper_model_definition['JasperEncoder']['jasper'],
activation=jasper_model_definition['JasperEncoder']['activation'],
feat_in=jasper_model_definition['AudioToMelSpectrogramPreprocessor']['features'])
jasper_encoder.restore_from(CHECKPOINT_ENCODER, local_rank=0)
RuntimeError: Error(s) in loading state_dict for JasperEncoder:
Missing key(s) in state_dict: "encoder.0.mconv.0.conv.weight", "encoder.0.mconv.1.conv.weight", "encoder.0.mconv.2.weight", "encoder.0.mconv.2.bias", "encoder.0.mconv.2.running_mean", "encoder.0.mconv.2.running_var", "encoder.1.mconv.0.conv.weight", "encoder.1.mconv.1.conv.weight", "encoder.1.mconv.2.weight", "encoder.1.mconv.2.bias", "encoder.1.mconv.2.running_mean", "encoder.1.mconv.2.running_var", "encoder.1.mconv.5.conv.weight", "encoder.1.mconv.6.conv.weight", "encoder.1.mconv.7.weight", "encoder.1.mconv.7.bias", "encoder.1.mconv.7.running_mean", "encoder.1.mconv.7.running_var", "encoder.1.mconv.10.conv.weight", "encoder.1.mconv.11.conv.weight", "encoder.1.mconv.12.weight", "encoder.1.mconv.12.bias", "encoder.1.mconv.12.running_mean", "encoder.1.mconv.12.running_var", "encoder.1.mconv.15.conv.weight", "encoder.1.mconv.16.conv.weight", "encoder.1.mconv.17.weight", "encoder.1.mconv.17.bias", "encoder.1.mconv.17.running_mean", "encoder.1.mconv.17.running_var", "encoder.1.mconv.20.conv.weight", "encoder.1.mconv.21.conv.weight", "encoder.1.mconv.22.weight", "encoder.1.mconv.22.bias", "encoder.1.mconv.22.running_mean", "encoder.1.mconv.22.running_var", "encoder.1.res.0.0.conv.weight", "encoder.2.mconv.0.conv.weight", "encoder.2.mconv.1.conv.weight", "encoder.2.mconv.2.weight", "encoder.2.mconv.2.bias", "encoder.2.mconv.2.running_mean", "encoder.2.mconv.2.running_var", "encoder.2.mconv.5.conv.weight", "encoder.2.mconv.6.conv.weight", "encoder.2.mconv.7.weight", "encoder.2.m...
Unexpected key(s) in state_dict: "encoder.0.conv.0.weight", "encoder.0.conv.1.weight", "encoder.0.conv.2.weight", "encoder.0.conv.2.bias", "encoder.0.conv.2.running_mean", "encoder.0.conv.2.running_var", "encoder.0.conv.2.num_batches_tracked", "encoder.1.conv.0.weight", "encoder.1.conv.1.weight", "encoder.1.conv.2.weight", "encoder.1.conv.2.bias", "encoder.1.conv.2.running_mean", "encoder.1.conv.2.running_var", "encoder.1.conv.2.num_batches_tracked", "encoder.1.conv.5.weight", "encoder.1.conv.6.weight", "encoder.1.conv.7.weight", "encoder.1.conv.7.bias", "encoder.1.conv.7.running_mean", "encoder.1.conv.7.running_var", "encoder.1.conv.7.num_batches_tracked", "encoder.1.conv.10.weight", "encoder.1.conv.11.weight", "encoder.1.conv.12.weight", "encoder.1.conv.12.bias", "encoder.1.conv.12.running_mean", "encoder.1.conv.12.running_var", "encoder.1.conv.12.num_batches_tracked", "encoder.1.conv.15.weight", "encoder.1.conv.16.weight", "encoder.1.conv.17.weight", "encoder.1.conv.17.bias", "encoder.1.conv.17.running_mean", "encoder.1.conv.17.running_var", "encoder.1.conv.17.num_batches_tracked", "encoder.1.conv.20.weight", "encoder.1.conv.21.weight", "encoder.1.conv.22.weight", "encoder.1.conv.22.bias", "encoder.1.conv.22.running_mean", "encoder.1.conv.22.running_var", "encoder.1.conv.22.num_batches_tracked", "encoder.1.res.0.0.weight", "encoder.2.conv.0.weight", "encoder.2.conv.1.weight", "encoder.2.conv.2.weight", "encoder.2.conv.2.bias", "encoder.2.conv.2.running_mean", "encoder....
Is there anyway to make this work with the older models or get newer compatible pre-trained models? @okuchaiev