File size: 2,194 Bytes
c6b1960 |
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 |
import argparse
import functools
import os
from transformers import WhisperForConditionalGeneration, WhisperFeatureExtractor, WhisperTokenizerFast,\
WhisperProcessor
from peft import PeftModel, PeftConfig
from utils.utils import print_arguments, add_arguments
parser = argparse.ArgumentParser(description=__doc__)
add_arg = functools.partial(add_arguments, argparser=parser)
add_arg("lora_model", type=str, default="output/whisper-tiny/checkpoint-best/", help="微调保存的模型路径")
add_arg('output_dir', type=str, default='models/', help="合并模型的保存目录")
add_arg("local_files_only", type=bool, default=False, help="是否只在本地加载模型，不尝试下载")
args = parser.parse_args()
print_arguments(args)
assert os.path.exists(args.lora_model), f"模型文件{args.lora_model}不存在"
peft_config = PeftConfig.from_pretrained(args.lora_model)
#
base_model = WhisperForConditionalGeneration.from_pretrained(peft_config.base_model_name_or_path, device_map={"": "cpu"},
local_files_only=args.local_files_only)
model = PeftModel.from_pretrained(base_model, args.lora_model, local_files_only=args.local_files_only)
feature_extractor = WhisperFeatureExtractor.from_pretrained(peft_config.base_model_name_or_path,
local_files_only=args.local_files_only)
tokenizer = WhisperTokenizerFast.from_pretrained(peft_config.base_model_name_or_path,
local_files_only=args.local_files_only)
processor = WhisperProcessor.from_pretrained(peft_config.base_model_name_or_path,
local_files_only=args.local_files_only)
model = model.merge_and_unload()
model.train(False)
save_directory = os.path.join(args.output_dir, f'{os.path.basename(peft_config.base_model_name_or_path)}-finetune')
os.makedirs(save_directory, exist_ok=True)
model.save_pretrained(save_directory)
feature_extractor.save_pretrained(save_directory)
tokenizer.save_pretrained(save_directory)
processor.save_pretrained(save_directory)
print(f'合并模型保持在：{save_directory}') |