Peft库实战（二）：Lora微调mt5-xxl【Seq2Seq：翻译、自动摘要、信息抽取】

一、Lora微调mt5-xxl

GPU要求：至少 A100-SXM4-80GB * 2

batch_size：A100-SXM4-80GB * 2情况下最大 16

备注：mt5-xxl全参数微调，batch_size=2时，A100-SXM4-80GB至少需要5张

run_finetune_lora.py

import logging
import os
import sys
import numpy as np
from datasets import Dataset
from peft import PeftModel
from peft import LoraModel, LoraConfig, get_peft_model
import random
import torch
import json
import transformers
from transformers import AutoConfig, AutoTokenizer, AutoModelForSeq2SeqLM, DataCollatorForSeq2Seq, HfArgumentParser, Seq2SeqTrainer, Seq2SeqTrainingArguments
from transformers.trainer_utils import get_last_checkpoint, is_main_process
from arguments import ModelArguments, DataTrainingArguments
from utils import get_extract_metrics_f1


os.environ["WANDB_DISABLED"] = "true"
logger = logging.getLogger("__main__&#