书生XTuner微调个人小助手

创建目录用于同步资源

同步资源完成后，创建conda虚拟环境并命名如下

等待虚拟环境创建完成后激活环境

conda activate xtuner0121

安装依赖软件和依赖库

等待安装完成后执行命令安装依赖

# 创建一个目录，用来存放源代码
mkdir -p /root/InternLM/code

cd /root/InternLM/code

执行命令下载XTuner：git clone -b v0.1.21  https://github.com/InternLM/XTuner /root/InternLM/code/XTuner

进入目录，执行命令安装

最后，我们可以验证一下安装结果。

执行命令查看版本以验证安装：xtuner version

# 创建一个目录，用来存放微调的所有资料，后续的所有操作都在该路径中进行
mkdir -p /root/InternLM/XTuner

cd /root/InternLM/XTuner

mkdir -p Shanghai_AI_Laboratory

ln -s /root/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b Shanghai_AI_Laboratory/internlm2-chat-1_8b

查看目录结构

启动模型：

streamlit run /root/InternLM/Tutorial/tools/xtuner_streamlit_demo.py

微调前模型对话

# 导入库包
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
 
# 定义模型加载方法
def load_model(model_path):
    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
    model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.float16, trust_remote_code=True).cuda()
    model = model.eval()
    return tokenizer, model
 
# 定义对话方法
messages = []
 
def chat(input_text):
    length = 0
    for response, _ in model.stream_chat(tokenizer, input_text, messages):
        if response is not None:
            print(response[length:], flush=True, end="")
            length = len(response)
 
# 模型加载
tokenizer, model = load_model("/root/InternLM/XTuner/Shanghai_AI_Laboratory/internlm2-chat-1_8b")
 
# 对话
chat("请介绍一下你自己")
 
# 释放缓存
del tokenizer, model
 
torch.cuda.empty_cache()
 

配置名词解释

配置微调参数：

# Copyright (c) OpenMMLab. All rights reserved.
import torch
from datasets import load_dataset
from mmengine.dataset import DefaultSampler
from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
                            LoggerHook, ParamSchedulerHook)
from mmengine.optim import AmpOptimWrapper, CosineAnnealingLR, LinearLR
from peft import LoraConfig
from torch.optim import AdamW
from transformers import (AutoModelForCausalLM, AutoTokenizer,
                          BitsAndBytesConfig)

from xtuner.dataset import process_hf_dataset
from xtuner.dataset.collate_fns import default_collate_fn
from xtuner.dataset.map_fns import alpaca_map_fn, template_map_fn_factory
from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
                                 VarlenAttnArgsToMessageHubHook)
from xtuner.engine.runner import TrainLoop
from xtuner.model import SupervisedFinetune
from xtuner.parallel.sequence import SequenceParallelSampler
from xtuner.utils import PROMPT_TEMPLATE, SYSTEM_TEMPLATE

#######################################################################
#                          PART 1  Settings                           #
#######################################################################
# Model
pretrained_model_name_or_path = '/root/InternLM/XTuner/Shanghai_AI_Laboratory/internlm2-chat-1_8b'
use_varlen_attn = False

# Data
alpaca_en_path = 'datas/assistant.json'
prompt_template = PROMPT_TEMPLATE.internlm2_chat
max_length = 2048
pack_to_max_length = True

# parallel
sequence_parallel_size = 1

# Scheduler & Optimizer
batch_size = 1  # per_device
accumulative_counts = 16
accumulative_counts *= sequence_parallel_size
dataloader_num_workers = 0
max_epochs = 3
optim_type = AdamW
lr = 2e-4
betas = (0.9, 0.999)
weight_decay = 0
max_norm = 1  # grad clip
warmup_ratio = 0.03

# Save
save_steps = 500
save_total_limit = 2  # Maximum checkpoints to keep (-1 means unlimited)

# Evaluate the generation performance during the training
evaluation_freq = 500
SYSTEM = SYSTEM_TEMPLATE.alpaca
evaluation_inputs = [
    '请介绍一下你自己', 'Please introduce yourself'
]

#######################################################################
#                      PART 2  Model & Tokenizer                      #
#######################################################################
tokenizer = dict(
    type=AutoTokenizer.from_pretrained,
    pretrained_model_name_or_path=pretrained_model_name_or_path,
    trust_remote_code=True,
    padding_side='right')

model = dict(
    type=SupervisedFinetune,
    use_varlen_attn=use_varlen_attn,
    llm=dict(
        type=AutoModelForCausalLM.from_pretrained,
        pretrained_model_name_or_path=pretrained_model_name_or_path,
        trust_remote_code=True,
        torch_dtype=torch.float16,
        quantization_config=dict(
            type=BitsAndBytesConfig,
            load_in_4bit=True,
            load_in_8bit=False,
            llm_int8_threshold=6.0,
            llm_int8_has_fp16_weight=False,
            bnb_4bit_compute_dtype=torch.float16,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type='nf4')),
    lora=dict(
        type=LoraConfig,
        r=64,
        lora_alpha=16,
        lora_dropout=0.1,
        bias='none',
        task_type='CAUSAL_LM'))

#######################################################################
#                      PART 3  Dataset & Dataloader                   #
#######################################################################
alpaca_en = dict(
    type=process_hf_dataset,
    dataset=dict(type=load_dataset, path='json', data_files=dict(train=alpaca_en_path)),
    tokenizer=tokenizer,
    max_length=max_length,
    dataset_map_fn=None,
    template_map_fn=dict(
        type=template_map_fn_factory, template=prompt_template),
    remove_unused_columns=True,
    shuffle_before_pack=True,
    pack_to_max_length=pack_to_max_length,
    use_varlen_attn=use_varlen_attn)

sampler = SequenceParallelSampler \
    if sequence_parallel_size > 1 else DefaultSampler
train_dataloader = dict(
    batch_size=batch_size,
    num_workers=dataloader_num_workers,
    dataset=alpaca_en,
    sampler=dict(type=sampler, shuffle=True),
    collate_fn=dict(type=default_collate_fn, use_varlen_attn=use_varlen_attn))

#######################################################################
#                    PART 4  Scheduler & Optimizer                    #
#######################################################################
# optimizer
optim_wrapper = dict(
    type=AmpOptimWrapper,
    optimizer=dict(
        type=optim_type, lr=lr, betas=betas, weight_decay=weight_decay),
    clip_grad=dict(max_norm=max_norm, error_if_nonfinite=False),
    accumulative_counts=accumulative_counts,
    loss_scale='dynamic',
    dtype='float16')

# learning policy
# More information: https://github.com/open-mmlab/mmengine/blob/main/docs/en/tutorials/param_scheduler.md  # noqa: E501
param_scheduler = [
    dict(
        type=LinearLR,
        start_factor=1e-5,
        by_epoch=True,
        begin=0,
        end=warmup_ratio * max_epochs,
        convert_to_iter_based=True),
    dict(
        type=CosineAnnealingLR,
        eta_min=0.0,
        by_epoch=True,
        begin=warmup_ratio * max_epochs,
        end=max_epochs,
        convert_to_iter_based=True)
]

# train, val, test setting
train_cfg = dict(type=TrainLoop, max_epochs=max_epochs)

#######################################################################
#                           PART 5  Runtime                           #
#######################################################################
# Log the dialogue periodically during the training process, optional
custom_hooks = [
    dict(type=DatasetInfoHook, tokenizer=tokenizer),
    dict(
        type=EvaluateChatHook,
        tokenizer=tokenizer,
        every_n_iters=evaluation_freq,
        evaluation_inputs=evaluation_inputs,
        system=SYSTEM,
        prompt_template=prompt_template)
]

if use_varlen_attn:
    custom_hooks += [dict(type=VarlenAttnArgsToMessageHubHook)]

# configure default hooks
default_hooks = dict(
    # record the time of every iteration.
    timer=dict(type=IterTimerHook),
    # print log every 10 iterations.
    logger=dict(type=LoggerHook, log_metric_by_epoch=False, interval=10),
    # enable the parameter scheduler.
    param_scheduler=dict(type=ParamSchedulerHook),
    # save checkpoint per `save_steps`.
    checkpoint=dict(
        type=CheckpointHook,
        by_epoch=False,
        interval=save_steps,
        max_keep_ckpts=save_total_limit),
    # set sampler seed in distributed evrionment.
    sampler_seed=dict(type=DistSamplerSeedHook),
)

# configure environment
env_cfg = dict(
    # whether to enable cudnn benchmark
    cudnn_benchmark=False,
    # set multi process parameters
    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
    # set distributed parameters
    dist_cfg=dict(backend='nccl'),
)

# set visualizer
visualizer = None

# set log level
log_level = 'INFO'

# load from which checkpoint
load_from = None

# whether to resume training from the loaded checkpoint
resume = False

# Defaults to use random seed and disable `deterministic`
randomness = dict(seed=None, deterministic=False)

# set log processor
log_processor = dict(by_epoch=False)

开始微调训练

模型格式转换

模型转换的本质其实就是将原本使用 Pytorch 训练出来的模型权重文件转换为目前通用的 HuggingFace 格式文件，那么我们可以通过以下命令来实现一键转换。

我们可以使用 xtuner convert pth_to_hf 命令来进行模型格式转换。

xtuner convert pth_to_hf 命令用于进行模型格式转换。该命令需要三个参数：CONFIG 表示微调的配置文件， PATH_TO_PTH_MODEL 表示微调的模型权重文件路径，即要转换的模型权重， SAVE_PATH_TO_HF_MODEL 表示转换后的 HuggingFace 格式文件的保存路径。

除此之外，我们其实还可以在转换的命令中添加几个额外的参数，包括：

参数名	解释
--fp32	代表以fp32的精度开启，假如不输入则默认为fp16
--max-shard-size {GB}	代表每个权重文件最大的大小（默认为2GB）

模型合并

对于 LoRA 或者 QLoRA 微调出来的模型其实并不是一个完整的模型，而是一个额外的层（Adapter），训练完的这个层最终还是要与原模型进行合并才能被正常的使用。

对于全量微调的模型（full）其实是不需要进行整合这一步的，因为全量微调修改的是原模型的权重而非微调一个新的 Adapter ，因此是不需要进行模型整合的。

在 XTuner 中提供了一键合并的命令 xtuner convert merge，在使用前我们需要准备好三个路径，包括原模型的路径、训练好的 Adapter 层的（模型格式转换后的）路径以及最终保存的路径。

xtuner convert merge命令用于合并模型。该命令需要三个参数：LLM 表示原模型路径，ADAPTER 表示 Adapter 层的路径， SAVE_PATH 表示合并后的模型最终的保存路径。

在模型合并这一步还有其他很多的可选参数，包括：

参数名	解释
--max-shard-size {GB}	代表每个权重文件最大的大小（默认为2GB）
--device {device_name}	这里指的就是device的名称，可选择的有cuda、cpu和auto，默认为cuda即使用gpu进行运算
--is-clip	这个参数主要用于确定模型是不是CLIP模型，假如是的话就要加上，不是就不需要添加

export MKL_SERVICE_FORCE_INTEL=1
export MKL_THREADING_LAYER=GNU
xtuner convert merge /root/InternLM/XTuner/Shanghai_AI_Laboratory/internlm2-chat-1_8b ./hf ./merged --max-shard-size 2GB

微调后重启服务：