diff --git a/README.md b/README.md
index 6ccec8d..f4123c6 100755
--- a/README.md
+++ b/README.md
@@ -101,10 +101,10 @@ Training a VALL-E model is very, very meticulous. I've fiddled with a lot of """
 Both trained models *can* be exported, but is only required if loading them on systems without DeepSpeed for inferencing (Windows systems). To export the models, run:
 
 ```
-python -m vall_e.export ./models/ yaml=./config/custom.yml
+python -m vall_e.export yaml=./config/custom.yml
 ```
 
-This will export the latest checkpoint.
+This will export the latest checkpoints.
 
 ### Synthesis
 
diff --git a/vall_e/config.py b/vall_e/config.py
index 442b5f3..ff34c50 100755
--- a/vall_e/config.py
+++ b/vall_e/config.py
@@ -17,6 +17,8 @@ from pathlib import Path
 
 from omegaconf import OmegaConf
 
+from .utils.distributed import world_size
+
 @dataclass()
 class _Config:
 	cfg_path: str | None = None
@@ -401,7 +403,7 @@ class BitsAndBytes:
 @dataclass()
 class Config(_Config):
 	device: str = "cuda"
-	distributed: bool = False
+	#distributed: bool = False
 
 	dataset: Dataset = field(default_factory=lambda: Dataset)
 	models: Models = field(default_factory=lambda: Models)
@@ -415,6 +417,10 @@ class Config(_Config):
 	def sample_rate(self):
 		return 24_000
 
+	@property
+	def distributed(self):
+		return world_size() > 1
+
 	@cached_property
 	def get_spkr(self):
 		return eval(self.dataset.speaker_name_getter)
@@ -447,20 +453,21 @@ try:
 	cfg.bitsandbytes = BitsAndBytes(**cfg.bitsandbytes)
 
 	cfg.trainer.deepspeed = DeepSpeed(**cfg.trainer.deepspeed)
+	
+	# cached_property stopped working...
+	if cfg.dataset.use_hdf5:
+		try:
+			cfg.hdf5 = h5py.File(f'{cfg.cfg_path}/{cfg.dataset.hdf5_name}', 'r' if cfg.distributed else 'a')
+		except Exception as e:
+			print("Error while opening HDF5 file:", f'{cfg.cfg_path}/{cfg.dataset.hdf5_name}', str(e))
+			cfg.dataset.use_hdf5 = False
+
+	if not cfg.dataset.use_hdf5:
+		cfg.dataset.training = [ Path(dir) for dir in cfg.dataset.training ]
+		cfg.dataset.validation = [ Path(dir) for dir in cfg.dataset.validation ]
 except Exception as e:
 	pass
 
-# cached_property stopped working...
-if cfg.dataset.use_hdf5:
-	try:
-		cfg.hdf5 = h5py.File(f'{cfg.cfg_path}/{cfg.dataset.hdf5_name}', 'r' if cfg.distributed else 'a')
-	except Exception as e:
-		print("Error while opening HDF5 file:", f'{cfg.cfg_path}/{cfg.dataset.hdf5_name}', str(e))
-		cfg.dataset.use_hdf5 = False
-
-if not cfg.dataset.use_hdf5:
-	cfg.dataset.training = [ Path(dir) for dir in cfg.dataset.training ]
-	cfg.dataset.validation = [ Path(dir) for dir in cfg.dataset.validation ]
 
 if __name__ == "__main__":
 	print(cfg)
diff --git a/vall_e/export.py b/vall_e/export.py
index 093f143..8f51871 100755
--- a/vall_e/export.py
+++ b/vall_e/export.py
@@ -4,31 +4,22 @@ import torch
 
 from .data import get_phone_symmap
 from .train import load_engines
-
-def load_models():
-	models = {}
-	engines = load_engines()
-	for name in engines:
-		model = engines[name].module.cpu()
-		
-		model.phone_symmap = get_phone_symmap()
-
-		models[name] = model
-
-	return models
+from .config import cfg
 
 def main():
 	parser = argparse.ArgumentParser("Save trained model to path.")
-	parser.add_argument("path")
+	#parser.add_argument("--yaml", type=Path, default=None)
 	args = parser.parse_args()
 
-	models = load_models()
-	for name in models:
-		model = models[name]
-
-		outpath = f'{args.path}/{name}.pt'
+	engines = load_engines()
+	for name, engine in engines.items():
+		outpath = cfg.ckpt_dir / name / "fp32.pth"
 		torch.save({
-			'module': model.state_dict()
+			"global_step": engine.global_step,
+			"micro_step": engine.micro_step,
+			'module': engine.module.to('cpu', dtype=torch.float32).state_dict(),
+			#'optimizer': engine.optimizer.state_dict(),
+			'symmap': get_phone_symmap(),
 		}, outpath)
 		print(f"Exported {name} to {outpath}")
 
diff --git a/vall_e/inference.py b/vall_e/inference.py
index e6c8193..9983934 100755
--- a/vall_e/inference.py
+++ b/vall_e/inference.py
@@ -19,6 +19,9 @@ class TTS():
 
 		self.input_sample_rate = 24000
 		self.output_sample_rate = 24000
+
+		if config:
+			cfg.load_yaml( config )
 		
 		if ar_ckpt and nar_ckpt:
 			self.ar_ckpt = ar_ckpt
@@ -33,7 +36,7 @@ class TTS():
 					self.nar = model.to(self.device, dtype=torch.float32)
 					self.nar.load_state_dict(torch.load(self.nar_ckpt)['module'])
 		else:
-			self.load_models( config )
+			self.load_models()
 
 		self.symmap = get_phone_symmap()
 		self.ar.eval()
@@ -41,10 +44,7 @@ class TTS():
 
 		self.loading = False 
 
-	def load_models( self, config_path ):
-		if config_path:
-			cfg.load_yaml( config_path )
-
+	def load_models( self ):
 		print("Loading models...")
 		models = load_models()
 		print("Loaded models")
diff --git a/vall_e/utils/distributed.py b/vall_e/utils/distributed.py
index e80b0dd..e43c7e5 100755
--- a/vall_e/utils/distributed.py
+++ b/vall_e/utils/distributed.py
@@ -48,6 +48,9 @@ def local_rank():
 def global_rank():
 	return int(os.getenv("RANK", 0))
 
+def world_size():
+	return int(os.getenv("WORLD_SIZE", 1))
+
 
 def is_local_leader():
 	return local_rank() == 0