File size: 1,619 Bytes
3a5a873 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 |
# you have got to be shitting me
import huggingface_hub
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch
import os
import argparse
parser = argparse.ArgumentParser(
prog='loraize',
description='Apply one or more loras to a model, and then save it',
epilog='BOTTOM TEXT')
parser.add_argument(
'model',
type=str,
help='path or HF name of a base model',
)
parser.add_argument(
'lora',
type=str,
help='one or more LORAs to apply',
nargs='+')
parser.add_argument(
'output_dir',
type=str,
help='output directory',
)
args = parser.parse_args()
print(f"Loading bassoon model:", args.model)
base_model = AutoModelForCausalLM.from_pretrained(
args.model,
return_dict=True,
torch_dtype=torch.bfloat16,
device_map="cpu",
)
for lora in args.lora:
print(f"Loading LORA: ",lora)
model = PeftModel.from_pretrained(
base_model,
lora,
device_map="cpu"
)
print(f"Good luck, bitches. Unloading.")
print("This gon' take a sec.")
model = model.merge_and_unload()
tokenizer = AutoTokenizer.from_pretrained(args.model)
model.save_pretrained(args.output_dir, safe_serialization=True, max_shard_size='10GB')
tokenizer.save_pretrained(args.output_dir)
|