text-generation-webui/modules/LoRA.py
2023-03-23 16:49:41 -03:00

31 lines
1.1 KiB
Python

from pathlib import Path
import modules.shared as shared
from modules.models import load_model
def add_lora_to_model(lora_name):
from peft import PeftModel
# Is there a more efficient way of returning to the base model?
if lora_name == "None":
print("Reloading the model to remove the LoRA...")
shared.model, shared.tokenizer = load_model(shared.model_name)
else:
print(f"Adding the LoRA {lora_name} to the model...")
params = {}
if not shared.args.cpu:
params['dtype'] = shared.model.dtype
if hasattr(shared.model, "hf_device_map"):
params['device_map'] = {"base_model.model."+k: v for k, v in shared.model.hf_device_map.items()}
elif shared.args.load_in_8bit:
params['device_map'] = {'': 0}
shared.model = PeftModel.from_pretrained(shared.model, Path(f"loras/{lora_name}"), **params)
if not shared.args.load_in_8bit and not shared.args.cpu:
shared.model.half()
if not hasattr(shared.model, "hf_device_map"):
shared.model.cuda()