Parcourir la source

add batch interval for saving model

nichongjia-2007 il y a 3 ans
Parent
commit
37c45ee8d7
2 fichiers modifiés avec 31 ajouts et 5 suppressions
  1. 6 0
      funasr/tasks/asr.py
  2. 25 5
      funasr/train/trainer.py

+ 6 - 0
funasr/tasks/asr.py

@@ -412,6 +412,12 @@ class ASRTask(AbsTask):
             default="13_15",
             help="The range of noise decibel level.",
         )
+        parser.add_argument(
+            "--batch_interval",
+            type=int,
+            default=10000,
+            help="The batch interval for saving model.",
+        )
 
         for class_choices in cls.class_choices_list:
             # Append --<name> and --<name>_conf.

+ 25 - 5
funasr/train/trainer.py

@@ -94,7 +94,7 @@ class TrainerOptions:
     wandb_model_log_interval: int
     use_pai: bool
     oss_bucket: Union[oss2.Bucket, None]
-
+    batch_interval: int
 
 class Trainer:
     """Trainer having a optimizer.
@@ -186,7 +186,10 @@ class Trainer:
                 logging.warning("No keep_nbest_models is given. Change to [1]")
                 trainer_options.keep_nbest_models = [1]
             keep_nbest_models = trainer_options.keep_nbest_models
-
+     
+        #assert batch_interval is set and >0
+        assert trainer_options.batch_interval > 0
+ 
         output_dir = Path(trainer_options.output_dir)
         reporter = Reporter()
         if trainer_options.use_amp:
@@ -560,13 +563,30 @@ class Trainer:
         # [For distributed] Because iteration counts are not always equals between
         # processes, send stop-flag to the other processes if iterator is finished
         iterator_stop = torch.tensor(0).to("cuda" if ngpu > 0 else "cpu")
-
+        
+        #get the rank
+        rank = distributed_option.dist_rank
+        #get the num batch updates
+        num_batch_updates = 0
+        #ouput dir
+        output_dir = Path(options.output_dir)
+        #batch interval
+        batch_interval = options.batch_interval       
+        assert batch_interval > 0
+ 
         start_time = time.perf_counter()
         for iiter, (_, batch) in enumerate(
             reporter.measure_iter_time(iterator, "iter_time"), 1
         ):
             assert isinstance(batch, dict), type(batch)
-
+        
+            if rank == 0 and hasattr(model.module, "num_updates"):
+                num_batch_updates = model.module.get_num_updates()
+                if (num_batch_updates%batch_interval == 0) and (options.oss_bucket is not None) and options.use_pai:
+                    buffer = BytesIO()
+                    torch.save(model.state_dict(), buffer)
+                    options.oss_bucket.put_object(os.path.join(output_dir, f"{num_batch_updates}batch.pth"), buffer.getvalue())
+ 
             if distributed:
                 torch.distributed.all_reduce(iterator_stop, ReduceOp.SUM)
                 if iterator_stop > 0:
@@ -811,4 +831,4 @@ class Trainer:
         else:
             if distributed:
                 iterator_stop.fill_(1)
-                torch.distributed.all_reduce(iterator_stop, ReduceOp.SUM)
+                torch.distributed.all_reduce(iterator_stop, ReduceOp.SUM)