diff --git a/egs/librispeech/ASR/pruned_transducer_stateless8/train.py b/egs/librispeech/ASR/pruned_transducer_stateless8/train.py index 41adda012..bee414292 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless8/train.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless8/train.py @@ -1127,7 +1127,16 @@ def run(rank, world_size, args): logging.info("Using DDP") model = DDP(model, device_ids=[rank], find_unused_parameters=True) - optimizer = ScaledAdam(model.parameters(), lr=params.base_lr, clipping_scale=2.0) + parameters_names = [] + parameters_names.append( + [name_param_pair[0] for name_param_pair in model.named_parameters()] + ) + optimizer = ScaledAdam( + model.parameters(), + lr=params.base_lr, + clipping_scale=2.0, + parameters_names=parameters_names, + ) scheduler = Eden(optimizer, params.lr_batches, params.lr_epochs)