Merge pull request #143 from NanoCode012/fix/deprecate-prepare-8bit-training

src/axolotl/utils/models.py

@@ -128,7 +128,8 @@ def load_model(
             )
 
             replace_peft_model_with_int4_lora_model()
-        from peft import prepare_model_for_int8_training
+        else:
+            from peft import prepare_model_for_kbit_training
     except Exception as err:
         logging.exception(err)
         raise err
@@ -269,8 +270,8 @@ def load_model(
         (cfg.adapter == "lora" and load_in_8bit)
         or (cfg.adapter == "qlora" and cfg.load_in_4bit)
     ):
-        logging.info("converting PEFT model w/ prepare_model_for_int8_training")
-        model = prepare_model_for_int8_training(model)
+        logging.info("converting PEFT model w/ prepare_model_for_kbit_training")
+        model = prepare_model_for_kbit_training(model)
 
     model, lora_config = load_adapter(model, cfg, adapter)