Exclude size=500 dim from projection; try to use double for model average

egs/librispeech/ASR/pruned_transducer_stateless4/optim.py

@@ -671,7 +671,9 @@ class Cain(Optimizer):
             # see for each dim in turn whether we want to perform any changes in co-ordinates,
             # or store any stats.
             size = grad.shape[dim]
-            if size <= 3 or size % 2 == 1 or size >= 2048 or size == numel:
+            if size <= 3 or size % 2 == 1 or size == 500 or size >= 2048 or size == numel:
+                # 500: exclude embedding dim, will later find a better way to do this.
+
                 # we don't do any such co-ordinate changes in dims with sizes
                 # that are too small (no point) or large (too slow), or that are
                 # assumed convolutional (because they are odd).  We can revisit

egs/librispeech/ASR/pruned_transducer_stateless4/train.py

@@ -274,6 +274,19 @@ def get_parser():
         """,
     )
 
+    parser.add_argument(
+        "--average-decay",
+        type=int,
+        default=100,
+        help="""Update the averaged model, namely `model_avg`, after processing
+        this number of batches. `model_avg` is a separate version of model,
+        in which each floating-point parameter is the average of all the
+        parameters from the start of training. Each time we take the average,
+        we do: `model_avg = model * (average_period / batch_idx_train) +
+            model_avg * ((batch_idx_train - average_period) / batch_idx_train)`.
+        """,
+    )
+
     parser.add_argument(
         "--use-fp16",
         type=str2bool,
@@ -736,6 +749,7 @@ def train_one_epoch(
                 params=params,
                 model_cur=model,
                 model_avg=model_avg,
+
             )
 
         if (
@@ -859,7 +873,7 @@ def run(rank, world_size, args):
     model_avg: Optional[nn.Module] = None
     if rank == 0:
         # model_avg is only used with rank 0
-        model_avg = copy.deepcopy(model)
+        model_avg = copy.deepcopy(model).to(torch.float64)
 
     assert params.start_epoch > 0, params.start_epoch
     checkpoints = load_checkpoint_if_available(

icefall/checkpoint.py

@@ -86,7 +86,7 @@ def save_checkpoint(
     }
 
     if model_avg is not None:
-        checkpoint["model_avg"] = model_avg.state_dict()
+        checkpoint["model_avg"] = model_avg.to(torch.float32).state_dict()
 
     if params:
         for k, v in params.items():