Add LoRA support

modules/models.py

@@ -11,6 +11,8 @@ from accelerate import infer_auto_device_map, init_empty_weights
 from transformers import (AutoConfig, AutoModelForCausalLM, AutoTokenizer,
                           BitsAndBytesConfig)
 
+from peft import PeftModel
+
 import modules.shared as shared
 
 transformers.logging.set_verbosity_error()

modules/shared.py

@@ -2,7 +2,8 @@ import argparse
 
 model = None
 tokenizer = None
-model_name = ""
+model_name = "None"
+lora_name = "None"
 soft_prompt_tensor = None
 soft_prompt = False
 is_RWKV = False