fixing layer_replication error

handler.py

@@ -18,7 +18,7 @@ class EndpointHandler:
 			bnb_4bit_compute_dtype=torch.float16,
 		)
 		
-		# Load the model with quantization
+		# Load the model without extra arguments like `layer_replication`
 		self.model = AutoModelForCausalLM.from_pretrained(
 			config.base_model_name_or_path,
 			return_dict=True,

requirements.txt

@@ -1,4 +1,4 @@
 torch
 transformers
 optimum
-peft
+peft>=0.3.0