from local

egs/librispeech/ASR/incremental_transf/identity_train.py

@@ -982,6 +982,11 @@ def run(rank, world_size, args):
     transducer_model.load_state_dict(pre_trained_model, strict=True)
 
     model = get_interformer_model(transducer_model.encoder, params)
+    for n, p in model.named_parameters():
+        if 'pt_encoder' in n:
+            p.requires_grad = False
+        else:
+            print(n)
     '''
     for n, p in model.named_parameters():
         if 'layer' not in n:
@@ -1016,7 +1021,7 @@ def run(rank, world_size, args):
     if world_size > 1:
         logging.info("Using DDP")
         model = DDP(model, device_ids=[rank])
-
+    
     optimizer = Eve(model.parameters(), lr=params.initial_lr)
 
     scheduler = Eden(optimizer, params.lr_batches, params.lr_epochs)