adapt for S, M and L training subset

egs/wenetspeech/ASR/pruned_transducer_stateless2/asr_datamodule.py

@@ -198,6 +198,13 @@ class WenetSpeechAsrDataModule:
             help="lazily open CutSets to avoid OOM (for L|XL subset)",
         )
 
+        group.add_argument(
+            "--training-subset",
+            type=str,
+            default="L",
+            help="The training subset for using",
+        )
+
     def train_dataloaders(
         self,
         cuts_train: CutSet,
@@ -316,6 +323,7 @@ class WenetSpeechAsrDataModule:
 
         if sampler_state_dict is not None:
             logging.info("Loading sampler state dict")
+            print(sampler_state_dict.keys())
             train_sampler.load_state_dict(sampler_state_dict)
 
         # 'seed' is derived from the current random state, which will have
@@ -396,7 +404,6 @@ class WenetSpeechAsrDataModule:
             world_size=1,
             shuffle=False,
         )
-
         from lhotse.dataset.iterable_dataset import IterableDatasetWrapper
 
         test_iter_dataset = IterableDatasetWrapper(
@@ -417,14 +424,12 @@ class WenetSpeechAsrDataModule:
             logging.info("use lazy cuts")
             cuts_train = CutSet.from_jsonl_lazy(
                 self.args.manifest_dir
-                / "cuts_L.jsonl.gz"
+                / f"cuts_{self.args.training_subset}.jsonl.gz"
-                # use cuts_L_50_pieces.jsonl.gz for original experiments
             )
         else:
             cuts_train = CutSet.from_file(
                 self.args.manifest_dir
-                / "cuts_L.jsonl.gz"
+                / f"cuts_{self.args.training_subset}.jsonl.gz"
-                # use cuts_L_50_pieces.jsonl.gz for original experiments
             )
         return cuts_train
 

egs/wenetspeech/ASR/pruned_transducer_stateless2/decode.py

@@ -101,6 +101,15 @@ def get_parser():
         help="It specifies the checkpoint to use for decoding."
         "Note: Epoch counts from 0.",
     )
+
+    parser.add_argument(
+        "--batch",
+        type=int,
+        default=None,
+        help="It specifies the batch checkpoint to use for decoding."
+        "Note: Epoch counts from 0.",
+    )
+
     parser.add_argument(
         "--avg",
         type=int,
@@ -499,6 +508,11 @@ def main():
         model.load_state_dict(average_checkpoints(filenames, device=device))
     elif params.avg == 1:
         load_checkpoint(f"{params.exp_dir}/epoch-{params.epoch}.pt", model)
+    elif params.batch is not None:
+        filenames = f"{params.exp_dir}/checkpoint-{params.batch}.pt"
+        logging.info(f"averaging {filenames}")
+        model.to(device)
+        model.load_state_dict(average_checkpoints([filenames], device=device))
     else:
         start = params.epoch - params.avg + 1
         filenames = []

egs/wenetspeech/ASR/pruned_transducer_stateless2/train.py

@@ -251,7 +251,7 @@ def get_parser():
     parser.add_argument(
         "--save-every-n",
         type=int,
-        default=200,
+        default=8000,
         help="""Save checkpoint after processing this number of batches"
         periodically. We save checkpoint to exp-dir/ whenever
         params.batch_idx_train % save_every_n == 0. The checkpoint filename
@@ -279,6 +279,26 @@ def get_parser():
         help="Whether to use half precision training.",
     )
 
+    parser.add_argument(
+        "--valid-interval",
+        type=int,
+        default=3000,
+        help="""When training_subset is L, set the valid_interval to 3000.
+        When training_subset is M, set the valid_interval to 1000.
+        When training_subset is S, set the valid_interval to 400.
+        """,
+    )
+
+    parser.add_argument(
+        "--model-warm-step",
+        type=int,
+        default=3000,
+        help="""When training_subset is L, set the model_warm_step to 3000.
+        When training_subset is M, set the model_warm_step to 500.
+        When training_subset is S, set the model_warm_step to 100.
+        """,
+    )
+
     return parser
 
 
@@ -333,9 +353,8 @@ def get_params() -> AttributeDict:
             "best_train_epoch": -1,
             "best_valid_epoch": -1,
             "batch_idx_train": 0,
-            "log_interval": 1,
+            "log_interval": 50,
             "reset_interval": 200,
-            "valid_interval": 3000,
             # parameters for conformer
             "feature_dim": 80,
             "subsampling_factor": 4,
@@ -348,7 +367,6 @@ def get_params() -> AttributeDict:
             # parameters for joiner
             "joiner_dim": 512,
             # parameters for Noam
-            "model_warm_step": 3000,  # arg given to model, not for lrate
             "env_info": get_env_info(),
         }
     )