fix bug and remove bnb

autograd_4bit.py

@@ -249,7 +249,7 @@ def load_llama_model_4bit_low_ram_and_offload(config_path, model_path, lora_path
     if lora_path is not None:
         from peft import PeftModel
         from monkeypatch.peft_tuners_lora_monkey_patch import Linear4bitLt
-        model = PeftModel.from_pretrained(model, lora_path, device_map={'': 'cpu'}, torch_dtype=torch.float32)
+        model = PeftModel.from_pretrained(model, lora_path, device_map={'': 'cpu'}, torch_dtype=torch.float32, is_trainable=True)
         print(Style.BRIGHT + Fore.GREEN + '{} Lora Applied.'.format(lora_path))
 
     model.seqlen = seqlen

monkeypatch/peft_tuners_lora_monkey_patch.py

@@ -2,7 +2,6 @@ import math
 import re
 import torch
 import warnings
-import bitsandbytes as bnb
 
 from peft.tuners import lora
 from peft.tuners.lora import is_bnb_available, Linear, Linear8bitLt, LoraLayer

requirements.txt

@@ -1,6 +1,5 @@
 torch
 accelerate
-bitsandbytes
 datasets
 sentencepiece
 safetensors