[to #42322933] bug fix: deadlock when setting the thread number up to 90

Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/10743508

* fix: load model directly from .pth

modelscope/models/audio/kws/farfield/model.py

@@ -54,7 +54,8 @@ class FSMNSeleNetV2Decorator(TorchModel):
                 )
 
     def __del__(self):
-        self.tmp_dir.cleanup()
+        if hasattr(self, 'tmp_dir'):
+            self.tmp_dir.cleanup()
 
     def forward(self, input: Dict[str, Tensor]) -> Dict[str, Tensor]:
         return self.model.forward(input)

modelscope/msdatasets/task_datasets/audio/kws_farfield_dataset.py

@@ -188,11 +188,13 @@ class Worker(threading.Thread):
 
 
 class KWSDataLoader:
-    """
-    dataset:            the dataset reference
-    batchsize:          data batch size
-    numworkers:         no. of workers
-    prefetch:           prefetch factor
+    """ Load and organize audio data with multiple threads
+
+    Args:
+        dataset:            the dataset reference
+        batchsize:          data batch size
+        numworkers:         no. of workers
+        prefetch:           prefetch factor
     """
 
     def __init__(self, dataset, batchsize, numworkers, prefetch=2):
@@ -202,7 +204,7 @@ class KWSDataLoader:
         self.isrun = True
 
         # data queue
-        self.pool = queue.Queue(batchsize * prefetch)
+        self.pool = queue.Queue(numworkers * prefetch)
 
         # initialize workers
         self.workerlist = []
@@ -270,11 +272,11 @@ class KWSDataLoader:
             w.stopWorker()
 
         while not self.pool.empty():
-            self.pool.get(block=True, timeout=0.001)
+            self.pool.get(block=True, timeout=0.01)
 
         # wait workers terminated
         for w in self.workerlist:
             while not self.pool.empty():
-                self.pool.get(block=True, timeout=0.001)
+                self.pool.get(block=True, timeout=0.01)
             w.join()
         logger.info('KWSDataLoader: All worker stopped.')

modelscope/trainers/audio/kws_farfield_trainer.py

@@ -117,8 +117,7 @@ class KWSFarfieldTrainer(BaseTrainer):
         self._batch_size = dataloader_config.batch_size_per_gpu
         if 'model_bin' in kwargs:
             model_bin_file = os.path.join(self.model_dir, kwargs['model_bin'])
-            checkpoint = torch.load(model_bin_file)
-            self.model.load_state_dict(checkpoint)
+            self.model = torch.load(model_bin_file)
         # build corresponding optimizer and loss function
         lr = self.cfg.train.optimizer.lr
         self.optimizer = optim.Adam(self.model.parameters(), lr)
@@ -219,7 +218,9 @@ class KWSFarfieldTrainer(BaseTrainer):
             # check point
             ckpt_name = 'checkpoint_{:04d}_loss_train_{:.4f}_loss_val_{:.4f}.pth'.format(
                 self._current_epoch, loss_train_epoch, loss_val_epoch)
-            torch.save(self.model, os.path.join(self.work_dir, ckpt_name))
+            save_path = os.path.join(self.work_dir, ckpt_name)
+            logger.info(f'Save model to {save_path}')
+            torch.save(self.model, save_path)
             # time spent per epoch
             epochtime = datetime.datetime.now() - epochtime
             logger.info('Epoch {:04d} time spent: {:.2f} hours'.format(

modelscope/utils/audio/audio_utils.py

@@ -43,7 +43,10 @@ def update_conf(origin_config_file, new_config_file, conf_item: [str, str]):
     def repl(matched):
         key = matched.group(1)
         if key in conf_item:
-            return conf_item[key]
+            value = conf_item[key]
+            if not isinstance(value, str):
+                value = str(value)
+            return value
         else:
             return None