experimental multi-gpu training (Linux only, because I can't into batch files)

.gitignore

@@ -1,7 +1,8 @@
 # ignores user files
-/tortoise-venv/
-/tortoise/voices/
-/models/
+/venv/
+/voices/*
+/models/*
+/training/*
 /config/*
 
 # Byte-compiled / optimized / DLL files

src/train.py

@@ -18,9 +18,12 @@ if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument('-opt', type=str, help='Path to option YAML file.', default='../options/train_vit_latent.yml', nargs='+') # ugh
     parser.add_argument('--launcher', choices=['none', 'pytorch'], default='none', help='job launcher')
+    parser.add_argument('--local_rank', type=int, help='Rank Number') 
     args = parser.parse_args()
     args.opt = " ".join(args.opt) # absolutely disgusting
 
+    os.environ['LOCAL_RANK'] = str(args.local_rank)
+
     with open(args.opt, 'r') as file:
         opt_config = yaml.safe_load(file)
 
@@ -71,7 +74,7 @@ def train(yaml, launcher='none'):
         print('Disabled distributed training.')
     else:
         opt['dist'] = True
-        init_dist('nccl')
+        tr.init_dist('nccl')
         trainer.world_size = torch.distributed.get_world_size()
         trainer.rank = torch.distributed.get_rank()
         torch.cuda.set_device(torch.distributed.get_rank())

src/utils.py

@@ -477,9 +477,7 @@ def compute_latents(voice, voice_latents_chunks, progress=gr.Progress(track_tqdm
 
 # superfluous, but it cleans up some things
 class TrainingState():
-	def __init__(self, config_path, keep_x_past_datasets=0, start=True):
-		self.cmd = ['train.bat', config_path] if os.name == "nt" else ['bash', './train.sh', config_path]
-
+	def __init__(self, config_path, keep_x_past_datasets=0, start=True, gpus=1):
 		# parse config to get its iteration
 		with open(config_path, 'r') as file:
 			self.config = yaml.safe_load(file)
@@ -530,9 +528,11 @@ class TrainingState():
 		if keep_x_past_datasets > 0:
 			self.cleanup_old(keep=keep_x_past_datasets)
 		if start:
-			self.spawn_process()
+			self.spawn_process(config_path=config_path, gpus=gpus)
+
+	def spawn_process(self, config_path, gpus=1):
+		self.cmd = ['train.bat', config_path] if os.name == "nt" else ['bash', './train.sh', str(int(gpus)), config_path]
 
-	def spawn_process(self):
 		print("Spawning process: ", " ".join(self.cmd))
 		self.process = subprocess.Popen(self.cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, universal_newlines=True)
 
@@ -745,7 +745,7 @@ class TrainingState():
 		if should_return:
 			return "".join(self.buffer)
 
-def run_training(config_path, verbose=False, buffer_size=8, keep_x_past_datasets=0, progress=gr.Progress(track_tqdm=True)):
+def run_training(config_path, verbose=False, gpus=1, buffer_size=8, keep_x_past_datasets=0, progress=gr.Progress(track_tqdm=True)):
 	global training_state
 	if training_state and training_state.process:
 		return "Training already in progress"
@@ -757,7 +757,7 @@ def run_training(config_path, verbose=False, buffer_size=8, keep_x_past_datasets
 	unload_whisper()
 	unload_voicefixer()
 
-	training_state = TrainingState(config_path=config_path, keep_x_past_datasets=keep_x_past_datasets)
+	training_state = TrainingState(config_path=config_path, keep_x_past_datasets=keep_x_past_datasets, gpus=gpus)
 
 	for line in iter(training_state.process.stdout.readline, ""):
 		
@@ -785,11 +785,13 @@ def update_training_dataplot(config_path=None):
 	update = None
 
 	if not training_state:
-		training_state = TrainingState(config_path=config_path, start=False)
-		update = gr.LinePlot.update(value=pd.DataFrame(training_state.losses))
-		del training_state
-		training_state = None
-	else:
+		if config_path:
+			training_state = TrainingState(config_path=config_path, start=False)
+			if training_state.losses:
+				update = gr.LinePlot.update(value=pd.DataFrame(training_state.losses))
+			del training_state
+			training_state = None
+	elif training_state.losses:
 		update = gr.LinePlot.update(value=pd.DataFrame(training_state.losses))
 
 	return update
@@ -1285,12 +1287,13 @@ def setup_args():
 	if not args.device_override:
 		set_device_name(args.device_override)
 
+
 	args.listen_host = None
 	args.listen_port = None
 	args.listen_path = None
 	if args.listen:
 		try:
-			match = re.findall(r"^(?:(.+?):(\d+))?(\/.+?)?$", args.listen)[0]
+			match = re.findall(r"^(?:(.+?):(\d+))?(\/.*?)?$", args.listen)[0]
 
 			args.listen_host = match[0] if match[0] != "" else "127.0.0.1"
 			args.listen_port = match[1] if match[1] != "" else None

src/webui.py

@@ -546,6 +546,7 @@ def setup_gradio():
 						verbose_training = gr.Checkbox(label="Verbose Console Output", value=True)
 						training_buffer_size = gr.Slider(label="Console Buffer Size", minimum=4, maximum=32, value=8)
 						training_keep_x_past_datasets = gr.Slider(label="Keep X Previous States", minimum=0, maximum=8, value=0, step=1)
+						training_gpu_count = gr.Number(label="GPUs", value=1)
 						with gr.Row():
 							start_training_button = gr.Button(value="Train")
 							stop_training_button = gr.Button(value="Stop")
@@ -751,6 +752,7 @@ def setup_gradio():
 			inputs=[
 				training_configs,
 				verbose_training,
+				training_gpu_count,
 				training_buffer_size,
 				training_keep_x_past_datasets,
 			],

train.sh

@@ -1,4 +1,13 @@
 #!/bin/bash
 source ./venv/bin/activate
-python3 ./src/train.py -opt "$1"
+
+GPUS=$1
+CONFIG=$2
+PORT=1234
+
+if (( $GPUS > 1 )); then
+	python3 -m torch.distributed.launch --nproc_per_node=$GPUS --master_port=$PORT ./src/train.py -opt "$CONFIG" --launcher=pytorch
+else
+	python3 ./src/train.py -opt "$CONFIG"
+fi
 deactivate