第二期书生浦语大模型实战营第四次课程笔记----XTuner 微调 LLM：1.8B、多模态、Agent

haidizym

已于 2024-04-17 15:55:27 修改

阅读量926

点赞数 22

分类专栏：大模型文章标签：笔记

于 2024-04-17 15:43:18 首次发布

本文链接：https://blog.csdn.net/haidizym/article/details/137870696

版权

大模型专栏收录该内容

15 篇文章 0 订阅

订阅专栏

1、微调理论讲解及 XTuner 介绍
2、XTuner 微调小助手个人认知实战
3、XTuner 微调 llava 图片理解多模态模型实战
【视频地址】： https://b23.tv/QUhT6ni
【课程文档】：https://github.com/InternLM/Tutorial/blob/camp2/xtuner/readme.md
【作业文档】：https://github.com/InternLM/Tutorial/blob/camp2/xtuner/homework.md

微调方式

在这里插入图片描述

还有agent微调哦

数据集格式：https://github.com/InternLM/xtuner/blob/main/docs/zh_cn/user_guides/dataset_format.md

微调工具：xtuner

在这里插入图片描述

#Xtuner
安装：pip install xtuner
直接训练：
挑选配置模板：xtuner list-cfg-p internlm_20b
一键训练：xtuner train internlm_20b_qlora_oasst1_512_e3
修改后训练：
拷贝配置模板： xtuner copy-cfg internlm_20b_qlora_oasst1_512_e3 ./
修改配置模板：vim internlm_20b_qlora_oasst1_512_e3_copy.py
启动训练：xtuner train internlm_20b_qlora_oasst1_512_e3_copy.py
训练后对话测试：
Float16模型对话：xtunerchat internlm/ internlm_20b
4bit模型对话：xtunerchat internlm/ internlm_20b --bits4
加载adpter模型对话：xtunerchat internlm/ internlm_20b --adapter $ADAPTER_DIR
加速训练：
xtuner train internlm_20b_qlora_oasst1_512_e3 --deepspeed deepspeed_zero3
在这里插入图片描述

单模态（文字）增量续训

教程：https://github.com/InternLM/xtuner/blob/main/docs/zh_cn/user_guides/incremental_pretraining.md
教程2：https://github.com/InternLM/InternEvo/blob/develop/doc/en/usage.md

installation

dataset preparation

[{
    "conversation":[
        {
            "input": "",
            "output": "xxx"
        },
    ]
}]

and model training configuration

from xtuner.dataset import process_hf_dataset
from datasets import load_dataset
- from xtuner.dataset.map_fns import oasst1_map_fn, template_map_fn_factory
+ from mmengine.config import read_base
+ with read_base():
+     from .map_fn import custom_map_fn
...
#######################################################################
#                          PART 1  Settings                           #
#######################################################################
- data_path = 'timdettmers/openassistant-guanaco'
- prompt_template = PROMPT_TEMPLATE.internlm_chat
+ data_path = 'path/to/your/data'
#######################################################################
#                      STEP 3  Dataset & Dataloader                   #
#######################################################################
train_dataset = dict(
    type=process_hf_dataset,
    dataset=dict(type=load_dataset, path=data_path),
    tokenizer=tokenizer,
    max_length=max_length,
-   dataset_map_fn=oasst1_map_fn,
+   dataset_map_fn=custom_map_fn,
-   template_map_fn=dict(
-       type=template_map_fn_factory, template=prompt_template),
+   template_map_fn=None,
    remove_unused_columns=True,
    shuffle_before_pack=True,
    pack_to_max_length=pack_to_max_length)
...
#######################################################################
#                           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,
-       instruction=prompt_template.INSTRUCTION)
]
...

单模态（文字）指令跟随微调

installation

操作平台：https://studio.intern-ai.org.cn/console/instance/

# 如果你是在 InternStudio 平台，则从本地 clone 一个已有 pytorch 的环境：
# pytorch    2.0.1   py3.10_cuda11.7_cudnn8.5.0_0
studio-conda xtuner0.1.17
# 如果你是在其他平台：
# conda create --name xtuner0.1.17 python=3.10 -y

# 激活环境
conda activate xtuner0.1.17
# 进入家目录 （~的意思是 “当前用户的home路径”）
cd ~
# 创建版本文件夹并进入，以跟随本教程
mkdir -p /root/xtuner0117 && cd /root/xtuner0117

# 拉取 0.1.17 的版本源码
git clone -b v0.1.17  https://github.com/InternLM/xtuner
# 无法访问github的用户请从 gitee 拉取:
# git clone -b v0.1.15 https://gitee.com/Internlm/xtuner

# 进入源码目录
cd /root/xtuner0117/xtuner

# 从源码安装 XTuner
pip install -e '.[all]'

dataset preparation

# 前半部分是创建一个文件夹，后半部分是进入该文件夹。
mkdir -p /root/ft && cd /root/ft

# 在ft这个文件夹里再创建一个存放数据的data文件夹
mkdir -p /root/ft/data && cd /root/ft/data

# 创建 `generate_data.py` 文件
touch /root/ft/data/generate_data.py

import json

# 设置用户的名字
name = '朱娅梅'
# 设置需要重复添加的数据次数
n =  10000

# 初始化OpenAI格式的数据结构
data = [
    {
        "messages": [
            {
                "role": "user",
                "content": "请做一下自我介绍"
            },
            {
                "role": "assistant",
                "content": "我是{}的小助手，内在是上海AI实验室书生·浦语的1.8B大模型哦".format(name)
            }
        ]
    }
]

# 通过循环，将初始化的对话数据重复添加到data列表中
for i in range(n):
    data.append(data[0])

# 将data列表中的数据写入到一个名为'personal_assistant.json'的文件中
with open('personal_assistant.json', 'w', encoding='utf-8') as f:
    # 使用json.dump方法将数据以JSON格式写入文件
    # ensure_ascii=False 确保中文字符正常显示
    # indent=4 使得文件内容格式化，便于阅读
    json.dump(data, f, ensure_ascii=False, indent=4)

# 确保先进入该文件夹
cd /root/ft/data

# 运行代码
python /root/ft/data/generate_data.py

and model training configuration

# 创建目标文件夹，确保它存在。
# -p选项意味着如果上级目录不存在也会一并创建，且如果目标文件夹已存在则不会报错。
mkdir -p /root/ft/model

# 复制内容到目标文件夹。-r选项表示递归复制整个文件夹。
cp -r /root/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b/* /root/ft/model/

# 删除/root/ft/model目录
rm -rf /root/ft/model

# 创建符号链接
ln -s /root/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b /root/ft/model

# 列出所有内置配置文件
# xtuner list-cfg

# 假如我们想找到 internlm2-1.8b 模型里支持的配置文件
xtuner list-cfg -p internlm2_1_8b

# 创建一个存放 config 文件的文件夹
mkdir -p /root/ft/config

# 使用 XTuner 中的 copy-cfg 功能将 config 文件复制到指定的位置
xtuner copy-cfg internlm2_1_8b_qlora_alpaca_e3 /root/ft/config

/root/ft/config/internlm2_1_8b_qlora_alpaca_e3_copy.py

# 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 openai_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/ft/model'
use_varlen_attn = False

# Data
alpaca_en_path = '/root/ft/data/personal_assistant.json'
prompt_template = PROMPT_TEMPLATE.default
max_length = 1024
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 = 2
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 = 300
save_total_limit = 3  # Maximum checkpoints to keep (-1 means unlimited)

# Evaluate the generation performance during the training
evaluation_freq = 300
SYSTEM = ''
evaluation_inputs = ['请你介绍一下你自己', '你是谁', '你是我的小助手吗']

#######################################################################
#                      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=openai_map_fn,
    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)

# 指定保存路径
xtuner train /root/ft/config/internlm2_1_8b_qlora_alpaca_e3_copy.py --work-dir /root/ft/train
# 使用 deepspeed 来加速训练
xtuner train /root/ft/config/internlm2_1_8b_qlora_alpaca_e3_copy.py --work-dir /root/ft/train_deepspeed --deepspeed deepspeed_zero2

模型转换、整合与部署

# 创建一个保存转换后 Huggingface 格式的文件夹
mkdir -p /root/ft/huggingface

# 模型转换
# xtuner convert pth_to_hf ${配置文件地址} ${权重文件地址} ${转换后模型保存地址}
xtuner convert pth_to_hf /root/ft/train/internlm2_1_8b_qlora_alpaca_e3_copy.py /root/ft/train/iter_768.pth /root/ft/huggingface

# 创建一个名为 final_model 的文件夹存储整合后的模型文件
mkdir -p /root/ft/final_model

# 解决一下线程冲突的 Bug 
export MKL_SERVICE_FORCE_INTEL=1

# 进行模型整合
# xtuner convert merge  ${NAME_OR_PATH_TO_LLM} ${NAME_OR_PATH_TO_ADAPTER} ${SAVE_PATH} 
xtuner convert merge /root/ft/model /root/ft/huggingface /root/ft/final_model

# 与模型进行对话
xtuner chat /root/ft/final_model --prompt-template internlm2_chat
# 同样的我们也可以和原模型进行对话进行对比
xtuner chat /root/ft/model --prompt-template internlm2_chat

多模态（识图）预训练/指令跟随微调

操作平台：https://studio.intern-ai.org.cn/console/instance/

installation

# 如果你是在 InternStudio 平台，则从本地 clone 一个已有 pytorch 的环境：
# pytorch    2.0.1   py3.10_cuda11.7_cudnn8.5.0_0

cd ~ && studio-conda xtuner0.1.17
# 如果你是在其他平台：
# conda create --name xtuner0.1.17 python=3.10 -y

# 激活环境
conda activate xtuner0.1.17
# 进入家目录 （~的意思是 “当前用户的home路径”）
cd ~
# 创建版本文件夹并进入，以跟随本教程
mkdir -p /root/xtuner0117 && cd /root/xtuner0117

# 拉取 0.1.17 的版本源码
git clone -b v0.1.17  https://github.com/InternLM/xtuner
# 无法访问github的用户请从 gitee 拉取:
# git clone -b v0.1.15 https://gitee.com/Internlm/xtuner

# 进入源码目录
cd /root/xtuner0117/xtuner

# 从源码安装 XTuner
pip install -e '.[all]' && cd ~

dataset preparation

[
    {
        "id": "随便什么字符串",
        "image": "图片文件的相对位置。相对谁？相对你后面config文件里指定的image_folder参数的路径。",
        "conversation": [
            {
                "from": "human",
                "value": "<image>\n第1个问题。"
            },
            {
                "from": "gpt",
                "value": "第1个回答"
            },
            {
                "from": "human",
                "value": "第2个问题。"
            },
            {
                "from": "gpt",
                "value": "第2个回答"
            },
            # ......
            {
                "from": "human",
                "value": "第n个问题。"
            },
            {
                "from": "gpt",
                "value": "第n个回答"
            },
        ]
    },

    # 下面是第2组训练数据了。

    {
        "id": "随便什么字符串",
        "image": "图片文件的相对位置。相对谁？相对你后面config文件里指定的image_folder参数的路径。",
        "conversation": [
            {
                "from": "human",
                "value": "<image>\n第1个问题。"
            },
            # ......
            {
                "from": "gpt",
                "value": "第n个回答"
            }
        ]
    }
]

cd ~ && git clone https://github.com/InternLM/tutorial -b camp2 && conda activate xtuner0.1.17 && cd tutorial

python /root/tutorial/xtuner/llava/llava_data/repeat.py \
  -i /root/tutorial/xtuner/llava/llava_data/unique_data.json \
  -o /root/tutorial/xtuner/llava/llava_data/repeated_data.json \
  -n 200

and model training configuration

#cp /root/tutorial/xtuner/llava/llava_data/internlm2_chat_1_8b_llava_tutorial_fool_config.py /root/tutorial/xtuner/llava/llava_internlm2_chat_1_8b_qlora_clip_vit_large_p14_336_lora_e1_gpu8_finetune_copy.py
# 查询xtuner内置配置文件
xtuner list-cfg -p llava_internlm2_chat_1_8b

# 拷贝配置文件到当前目录
xtuner copy-cfg \
  llava_internlm2_chat_1_8b_qlora_clip_vit_large_p14_336_lora_e1_gpu8_finetune \
  /root/tutorial/xtuner/llava

修改llava_internlm2_chat_1_8b_qlora_clip_vit_large_p14_336_lora_e1_gpu8_finetune_copy.py文件

# Model
- llm_name_or_path = 'internlm/internlm2-chat-1_8b'
+ llm_name_or_path = '/root/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b'
- visual_encoder_name_or_path = 'openai/clip-vit-large-patch14-336'
+ visual_encoder_name_or_path = '/root/share/new_models/openai/clip-vit-large-patch14-336'

# Specify the pretrained pth
- pretrained_pth = './work_dirs/llava_internlm2_chat_1_8b_clip_vit_large_p14_336_e1_gpu8_pretrain/iter_2181.pth'  # noqa: E501
+ pretrained_pth = '/root/share/new_models/xtuner/iter_2181.pth'

# Data
- data_root = './data/llava_data/'
+ data_root = '/root/tutorial/xtuner/llava/llava_data/'
- data_path = data_root + 'LLaVA-Instruct-150K/llava_v1_5_mix665k.json'
+ data_path = data_root + 'repeated_data.json'
- image_folder = data_root + 'llava_images'
+ image_folder = data_root

# Scheduler & Optimizer
- batch_size = 16  # per_device
+ batch_size = 1  # per_device


# evaluation_inputs
- evaluation_inputs = ['请描述一下这张图片','Please describe this picture']
+ evaluation_inputs = ['Please describe this picture','What is the equipment in the image?']

cd /root/tutorial/xtuner/llava/
xtuner train /root/tutorial/xtuner/llava/llava_internlm2_chat_1_8b_qlora_clip_vit_large_p14_336_lora_e1_gpu8_finetune_copy.py --deepspeed deepspeed_zero2

对比效果

Q1: Describe this image.
Q2: What is the equipment in the image?

微调前
加载 1.8B 和 Pretrain阶段产物(iter_2181) 到显存。

# 解决小bug
export MKL_SERVICE_FORCE_INTEL=1
export MKL_THREADING_LAYER=GNU

# pth转huggingface
xtuner convert pth_to_hf \
  llava_internlm2_chat_1_8b_clip_vit_large_p14_336_e1_gpu8_pretrain \
  /root/share/new_models/xtuner/iter_2181.pth \
  /root/tutorial/xtuner/llava/llava_data/iter_2181_hf

# 启动！
xtuner chat /root/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b \
  --visual-encoder /root/share/new_models/openai/clip-vit-large-patch14-336 \
  --llava /root/tutorial/xtuner/llava/llava_data/iter_2181_hf \
  --prompt-template internlm2_chat \
  --image /root/tutorial/xtuner/llava/llava_data/test_img/oph.jpg

微调后

# 解决小bug
export MKL_SERVICE_FORCE_INTEL=1
export MKL_THREADING_LAYER=GNU

# pth转huggingface
xtuner convert pth_to_hf \
  /root/tutorial/xtuner/llava/llava_internlm2_chat_1_8b_qlora_clip_vit_large_p14_336_lora_e1_gpu8_finetune_copy.py \
  /root/tutorial/xtuner/llava/work_dirs/llava_internlm2_chat_1_8b_qlora_clip_vit_large_p14_336_lora_e1_gpu8_finetune_copy/iter_1200.pth \
  /root/tutorial/xtuner/llava/llava_data/iter_1200_hf

# 启动！
xtuner chat /root/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b \
  --visual-encoder /root/share/new_models/openai/clip-vit-large-patch14-336 \
  --llava /root/tutorial/xtuner/llava/llava_data/iter_1200_hf \
  --prompt-template internlm2_chat \
  --image /root/tutorial/xtuner/llava/llava_data/test_img/oph.jpg