Update app.py

app.py

@@ -136,33 +136,25 @@ def load_model(model_name):
     tokenizer = AutoTokenizer.from_pretrained(model_name)
     tokenizer.pad_token = tokenizer.eos_token
 
-    # PEFT
-
-    # TODO: LoRA, QLoRA
-    
-    #peft_model = PeftModel.from_pretrained(
-    #    model,
-    #   adapter_name="lora",
-    #)
-
-    #print("### PEFT")
-    #peft_model.print_trainable_parameters() # trainable params: 6,815,744 || all params: 8,037,076,992 || trainable%: 0.0848
-    #print("###")
+    # PEFT, LoRA, QLoRA, see https://huggingface.co/blog/mlabonne/sft-llama3
 
     ###
     print("111")
     peft_config = LoraConfig(
-        lora_alpha=16,
-        lora_dropout=0,
         r=64,
-        bias="none",
-        task_type="CAUSAL_LM",
+        # TODO
+        #bias="none",
+        #lora_alpha=16,
+        #lora_dropout=0,
+        #task_type="CAUSAL_LM",
     )
     print("222")
     model = PeftModel.from_pretrained(base_model, "new_model", peft_config=peft_config)
     print("333")
     model = model.merge_and_unload()
     print("444")
+    model.print_trainable_parameters() # trainable params: 6,815,744 || all params: 8,037,076,992 || trainable%: 0.0848
+    print("555")
     ###
     
     return model, tokenizer