Added cfg.bitsandbytes.replace as a less intrusive alternative to cfg.bitsandbytes.inject to replace all Linear modules in a model

vall_e/config.py

@@ -303,7 +303,7 @@ class Models:
 class Hyperparameters:
 	batch_size: int = 8
 	gradient_accumulation_steps: int = 32
-	gradient_clipping: int = 100
+	gradient_clipping: int | float = 100
 
 	optimizer: str = "Adamw"
 	torch_optimizer: bool = False
@@ -532,6 +532,7 @@ class Inference:
 class BitsAndBytes:
 	enabled: bool = False
 	injects: bool = False
+	replace: bool = False
 
 	linear: bool = True
 	embedding: bool = True

vall_e/engines/__init__.py

@@ -44,6 +44,9 @@ def load_engines(training=True):
 		if inferencing:
 			model._cfg.training = False
 
+		if cfg.bitsandbytes.enabled and cfg.bitsandbytes.replace:
+			model.model = ml.replace_linear( model.model )
+
 		if backend == "local" or (backend == "deepspeed" and cfg.hyperparameters.torch_optimizer):
 			optimizer_class = None
 			params = {

vall_e/models/ar_nar.py

@@ -341,7 +341,7 @@ def example_usage():
 		'd_model': 1024, # 256, # 1024, # 1536
 		'n_heads': 16, # 4, # 16, # 24
 		'n_layers': 12, # 32
-		'n_experts': 8,
+		'n_experts': 1,
 	}
 	"""
 	kwargs = {
@@ -362,10 +362,13 @@ def example_usage():
 
 	model = AR_NAR(**kwargs).to(device)
 	steps = 500
-	#optimizer = ml.Prodigy(model.parameters(), lr=1.0)
-	optimizer = ml.AdamW(model.parameters(), lr=1.0e-4)
+	optimizer = ml.Prodigy(model.parameters(), lr=1.0)
+	#optimizer = ml.AdamW(model.parameters(), lr=1.0e-4)
 	engine = Engine(model=model, optimizer=optimizer)
 
+	if cfg.bitsandbytes.enabled and cfg.bitsandbytes.replace:
+		model.model = ml.replace_linear( model.model )
+
 	torch.save( {
 		'module': model.state_dict()
 	}, "./data/test.pth" )

vall_e/models/transformer.py

@@ -80,8 +80,8 @@ class Attention(nn.Module):
 		self.n_heads = n_heads
 		self.scale = dim_head**-0.5
 
-		self.to_qkv = nn.Linear(d_model, d_model * 3, bias=False)
-		self.to_out = nn.Linear(d_model, d_model)
+		self.to_qkv = ml.Linear(d_model, d_model * 3, bias=False)
+		self.to_out = ml.Linear(d_model, d_model)
 
 	def forward(self, x, m):
 		"""
@@ -169,10 +169,10 @@ class Block(nn.Sequential):
 		n_ff = d_model * 4 # 1024 * 4 = 4096 feed-forwards
 		self.ffn = PrenormResidual(
 			nn.Sequential(
-				nn.Linear(d_model, n_ff),
+				ml.Linear(d_model, n_ff),
 				nn.GELU(),
 				nn.Dropout(p_dropout),
-				nn.Linear(n_ff, d_model),
+				ml.Linear(n_ff, d_model),
 			),
 			d_model=d_model,
 			p_dropout=p_dropout,

vall_e/utils/wrapper.py

@@ -83,6 +83,32 @@ if cfg.bitsandbytes.injects and cfg.bitsandbytes.enabled:
 	torch.optim.AdamW = AdamW
 	torch.optim.SGD = SGD
 
+# disgusting kludge, but it works
+def replace_linear( model ):
+	device =  next(model.parameters()).device
+	linears = [k.split('.') for k, m in model.named_modules() if type(m).__name__ == 'Linear']
+	for *parent, k in linears:
+		name = '.'.join(parent)
+
+		# copy parameters
+		m = getattr(
+			model.get_submodule(name),
+			k
+		)
+
+		in_features = m.in_features
+		out_features = m.out_features
+		bias = False if cfg.bitsandbytes.bitnet else m.bias # errors out with BitNet
+
+		# overwright
+		setattr(
+			model.get_submodule(name),
+			k,
+			Linear( in_features=in_features, out_features=out_features, bias=bias ).to(device)
+		)
+
+	return model
+
 # https://github.com/konstmish/prodigy
 try:
 	from prodigyopt import Prodigy