Llama3-Tutorial（Llama 3 超级课堂） 学习笔记

课程资料

1. 1. 五一 Llama 3 超级课堂 | 第二节 Llama 3 微调个人小助手认知（XTuner版）_哔哩哔哩_bilibili
   2. SmartFlowAI/Llama3-Tutorial: Llama3-Tutorial（XTuner、LMDeploy、OpenCompass） (github.com)
   3. 开发机 (intern-ai.org.cn)
   4. Llama3-Tutorial/docs/hello_world.md at main · SmartFlowAI/Llama3-Tutorial (github.com)

1. 预先准备

   1. 注册InternStudio、创建开发机并启动，这里要选择 “资源配置”，指的是GPU的资源。
   2. 安装VSCODE
   3. 建立远程链接
   4. 由于代码经常更新，所以需要及时同步最新的代码，如果那段程序调试出现问题，可以先尝试更新远程代码库。

git branch -r  #查看分支

git checkout main

git pull origin main

第一节 本地 Web Demo 部署

#安装环境

conda create -n llama3 python=3.10

conda activate llama3

conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia

#通过软连接链接大模型文件。

ln -s /root/share/new_models/meta-llama/Meta-Llama-3-8B-Instruct ~/model/Meta-Llama-3-8B-Instruct

#运行，下面有两种方式， 建议采用quant模式，占用资源较少。在最小的“资源配置”的开发机上也可以跑。

streamlit run ~/Llama3-Tutorial/tools/internstudio_web_demo.py ~/model/Meta-Llama-3-8B-Instruct

streamlit run ~/Llama3-Tutorial/tools/internstudio_quant_web_demo.py ~/model/Meta-Llama-3-8B-Instruct  (quant模式)

第二节 Llama 3 微调个人小助手认知（XTuner版）

#通过gdata.py，创建微调的数据集。这里可以随便起个名字，例如：“南方蓝天”

cd ~/Llama3-Tutorial

python tools/gdata.py

conda activate llama3   （激活环境）

cd ~/Llama3-Tutorial

# 开始训练,使用 deepspeed 加速，A100 40G显存 耗时24分钟

xtuner train configs/assistant/llama3_8b_instruct_qlora_assistant.py --work-dir /root/llama3_pth

运行完成后的目录

# Adapter PTH 转 HF 格式

xtuner convert pth_to_hf /root/llama3_pth/llama3_8b_instruct_qlora_assistant.py \

  /root/llama3_pth/iter_500.pth \

  /root/llama3_hf_adapter

# 模型合并

export MKL_SERVICE_FORCE_INTEL=1

xtuner convert merge /root/model/Meta-Llama-3-8B-Instruct \

  /root/llama3_hf_adapter\

  /root/llama3_hf_merged Adapter

第三节 Llama 3 图片理解能力微调（XTuner+LLaVA）

启动环境

conda create -n llama3 python=3.10

conda activate llama3

conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia

#单独安装lfs

pip install lfs

#需要的 openai/clip-vit-large-patch14-336，权重，即 Visual Encoder 权重。

cd ~/model

ln -s /root/share/new_models/openai/clip-vit-large-patch14-336 .

#然后我们准备 Llava 将要用到的 Image Projector 部分权重。

ln -s /root/share/new_models/xtuner/llama3-llava-iter_2181.pth .

#数据准备

cd ~

git clone https://github.com/InternLM/tutorial -b camp2

python ~/tutorial/xtuner/llava/llava_data/repeat.py \

  -i ~/tutorial/xtuner/llava/llava_data/unique_data.json \

  -o ~/tutorial/xtuner/llava/llava_data/repeated_data.json \

  -n 200

 

  微调过程

  启动训练：（！！！这一步需要50M的显存，由于开发机的显存不够，所以无法完成。有些遗憾）

  xtuner train ~/Llama3-Tutorial/configs/llama3-llava/llava_llama3_8b_instruct_qlora_clip_vit_large_p14_336_lora_e1_finetune.py \

  --work-dir ~/llama3_llava_pth --deepspeed deepspeed_zero2

 

  转换为 HuggingFace 格式

  xtuner convert pth_to_hf ~/Llama3-Tutorial/configs/llama3-llava/llava_llama3_8b_instruct_qlora_clip_vit_large_p14_336_lora_e1_finetune.py \

  ~/model/llama3-llava-iter_2181.pth \

  ~/llama3_llava_pth/pretrain_iter_2181_hf

   

第四节 Llama 3 高效部署实践（LMDeploy版）

环境配置

# 如果你是InternStudio 可以直接使用

# studio-conda -t lmdeploy -o pytorch-2.1.2

# 初始化环境

conda create -n lmdeploy python=3.10

conda activate lmdeploy

conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia

安装lmdeploy最新版。

cd ~

pip install -U lmdeploy[all]

LMDeploy Chat CLI 工具

conda activate lmdeploy   #切换环境

lmdeploy chat /root/model/Meta-Llama-3-8B-Instruct

3. LMDeploy模型量化(lite)

本部分内容主要介绍如何对模型进行量化。主要包括 KV8量化和W4A16量化。

3.1 设置最大KV Cache缓存大小

模型在运行时，占用的显存可大致分为三部分：模型参数本身占用的显存、KV Cache占用的显存，以及中间运算结果占用的显存。LMDeploy的KV Cache管理器可以通过设置--cache-max-entry-count参数，控制KV缓存占用剩余显存的最大比例。默认的比例为0.8。

lmdeploy chat /root/model/Meta-Llama-3-8B-Instruct/

内存基本都使用了。

lmdeploy chat /root/model/Meta-Llama-3-8B-Instruct/ --cache-max-entry-count 0.5

用量降低了一些

lmdeploy chat /root/model/Meta-Llama-3-8B-Instruct/ --cache-max-entry-count 0.01

3.2 使用W4A16量化

lmdeploy lite auto_awq \

   /root/model/Meta-Llama-3-8B-Instruct \

  --calib-dataset 'ptb' \

  --calib-samples 128 \

  --calib-seqlen 1024 \

  --w-bits 4 \

  --w-group-size 128 \

  --work-dir /root/model/Meta-Llama-3-8B-Instruct_4bit

lmdeploy chat /root/model/Meta-Llama-3-8B-Instruct_4bit --model-format awq

lmdeploy chat /root/model/Meta-Llama-3-8B-Instruct_4bit --model-format awq --cache-max-entry-count 0.01

已经降到了6G的内存。

3.3 在线量化 KV

自 v0.4.0 起，LMDeploy KV 量化方式有原来的离线改为在线。并且，支持两种数值精度 int4、int8。量化方式为 per-head per-token 的非对称量化。它具备以下优势：

量化不需要校准数据集

kv int8 量化精度几乎无损，kv int4 量化精度在可接受范围之内

推理高效，在 llama2-7b 上加入 int8/int4 kv 量化，RPS 相较于 fp16 分别提升近 30% 和 40%

支持 volta 架构（sm70）及以上的所有显卡型号：V100、20系列、T4、30系列、40系列、A10、A100 等等 通过 LMDeploy 应用 kv 量化非常简单，只需要设定 quant_policy 参数。LMDeploy 规定 qant_policy=4表示 kv int4 量化，quant_policy=8 表示 kv int8 量化。

4. LMDeploy服务（serve）

4.1 启动API服务器

lmdeploy serve api_server \

    /root/model/Meta-Llama-3-8B-Instruct \

    --model-format hf \

    --quant-policy 0 \

    --server-name 0.0.0.0 \

    --server-port 23333 \

    --tp 1

设定SSH的转发

ssh -CNg -L 23333:127.0.0.1:23333 root@ssh.intern-ai.org.cn -p 46647

4.2 命令行客户端连接API服务器

在“4.1”中，我们在终端里新开了一个API服务器。 本节中，我们要新建一个命令行客户端去连接API服务器。首先通过VS Code新建一个终端： 激活conda环境

1， 命令行方式

conda activate lmdeploy

lmdeploy serve api_client http://localhost:23333

2 网页方式

conda activate lmdeploy

pip install gradio==3.50.2

lmdeploy serve gradio http://localhost:23333 \

    --server-name 0.0.0.0 \

    --server-port 6006

5. 推理速度

克隆仓库

cd ~

git clone https://github.com/InternLM/lmdeploy.git

下载测试数据（642M）

cd /root/lmdeploy

wget https://hf-mirror.com/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json

执行 benchmark 命令(如果你的显存较小，可以调低--cache-max-entry-count)

python benchmark/profile_throughput.py \

    ShareGPT_V3_unfiltered_cleaned_split.json \

    /root/model/Meta-Llama-3-8B-Instruct \

    --cache-max-entry-count 0.8 \

    --concurrency 256 \

    --model-format hf \

    --quant-policy 0 \

    --num-prompts 10000

我在这运行出了一个错误提示， 让后将/root/lmdeploy/benchmark/profile_throughput.py 中的286行的“ArgumentHelper.enable_prefix_caching(pt_group)”注释掉就可以了。

测试的时候开发机还没有跑满

第五节 Llama 3 Agent 体验微调（XTuner版）

2.1 环境配置

环境配置还是llama3

conda activate llama3

安装XTuner（完成之前的课程，已经已经装完了）

cd ~

git clone -b v0.1.18 https://github.com/InternLM/XTuner

cd XTuner

pip install -e .[all]

2.2 模型准备

2.3 数据集准备

由于 HuggingFace 上的 Agent-FLAN 数据集暂时无法被 XTuner 直接加载，因此我们首先要下载到本地，然后转换成 XTuner 直接可用的格式。

cd ~

cp -r /root/share/new_models/internlm/Agent-FLAN .

chmod -R 755 Agent-FLAN

2.4 微调启动

由于训练时间太长，我们也为大家准备好了已经训练好且转换为 HuggingFace 格式的权重，可以直接使用。路径位于 

/share/new_models/agent-flan/iter_2316_hf。

如果要使用已经训练好的权重，可以使用如下指令合并权重：

export MKL_SERVICE_FORCE_INTEL=1

xtuner convert merge /root/model/Meta-Llama-3-8B-Instruct \

    /share/new_models/agent-flan/iter_2316_hf \

    ~/llama3_agent_pth/merged

4. Lagent Web Demo（在教材的后面，但是要先使用的内容，建议教材可以提到前面）

pip install lagent

streamlit run ~/Llama3-Tutorial/tools/agent_web_demo.py /root/model/Meta-Llama-3-8B-Instruct

streamlit run ~/Llama3-Tutorial/tools/agent_web_demo.py /root/llama3_agent_pth/merged

启动后对比问题，这次使用的问题是“查找关于InternLM2的论文”

对比之前的结果，这次已经可以自动调用ArxivSearch接口了。

第六节 Llama 3 能力评测（OpenCompass 版）

这次继续使用llama3环境

conda activate llama3

安装 OpenCompass。简介一下，OpenCompass是上海人工智能实验室开源的大模型评测平台，它涵盖了学科、语言、知识、理解、推理等五大评测维度，可以全面评估大模型的能力。上海人工智能实验室 (shlab.org.cn)

cd ~

git clone https://github.com/open-compass/opencompass opencompass

cd opencompass

pip install -e .

数据准备

下载数据集到 data/ 处

wget https://github.com/open-compass/opencompass/releases/download/0.2.2.rc1/OpenCompassData-core-20240207.zip

unzip OpenCompassData-core-20240207.zip

下载速度实在是太慢了，后面没法操作了，所以实战部分就到此结束， 后续的只是课程的内容。

命令行快速评测

OpenCompass 预定义了许多模型和数据集的配置，你可以通过 工具 列出所有可用的模型和数据集配置。

# 列出所有配置

# python tools/list_configs.py

# 列出所有跟 llama (模型)及 ceval（数据集） 相关的配置

python tools/list_configs.py llama ceval

以 C-Eval_gen 为例：

python run.py --datasets ceval_gen --hf-path /root/model/Meta-Llama-3-8B-Instruct --tokenizer-path /root/model/Meta-Llama-3-8B-Instruct --tokenizer-kwargs padding_side='left' truncation='left' trust_remote_code=True --model-kwargs trust_remote_code=True device_map='auto' --max-seq-len 2048 --max-out-len 16 --batch-size 4 --num-gpus 1 --debug

命令解析

python run.py \

--datasets ceval_gen \

--hf-path /root/model/Meta-Llama-3-8B-Instruct \  # HuggingFace 模型路径

--tokenizer-path /root/model/Meta-Llama-3-8B-Instruct \  # HuggingFace tokenizer 路径（如果与模型路径相同，可以省略）

--tokenizer-kwargs padding_side='left' truncation='left' trust_remote_code=True \  # 构建 tokenizer 的参数

--model-kwargs device_map='auto' trust_remote_code=True \  # 构建模型的参数

--max-seq-len 2048 \  # 模型可以接受的最大序列长度

--max-out-len 16 \  # 生成的最大 token 数

--batch-size 4  \  # 批量大小

--num-gpus 1 \ # 运行模型所需的 GPU 数量

--debug

评测完成后，将会看到：

dataset                                         version    metric         mode      opencompass.models.huggingface.HuggingFace_meta-llama_Meta-Llama-3-8B-Instruct

----------------------------------------------  ---------  -------------  ------  --------------------------------------------------------------------------------

ceval-computer_network                          db9ce2     accuracy       gen                                                                                63.16

ceval-operating_system                          1c2571     accuracy       gen                                                                                63.16

ceval-computer_architecture                     a74dad     accuracy       gen                                                                                52.38

ceval-college_programming                       4ca32a     accuracy       gen                                                                                62.16

ceval-college_physics                           963fa8     accuracy       gen                                                                                42.11

ceval-college_chemistry                         e78857     accuracy       gen                                                                                29.17

ceval-advanced_mathematics                      ce03e2     accuracy       gen                                                                                42.11

ceval-probability_and_statistics                65e812     accuracy       gen                                                                                27.78

ceval-discrete_mathematics                      e894ae     accuracy       gen                                                                                25

ceval-electrical_engineer                       ae42b9     accuracy       gen                                                                                32.43

ceval-metrology_engineer                        ee34ea     accuracy       gen                                                                                62.5

ceval-high_school_mathematics                   1dc5bf     accuracy       gen                                                                                 5.56

ceval-high_school_physics                       adf25f     accuracy       gen                                                                                26.32

ceval-high_school_chemistry                     2ed27f     accuracy       gen                                                                                63.16

ceval-high_school_biology                       8e2b9a     accuracy       gen                                                                                36.84

ceval-middle_school_mathematics                 bee8d5     accuracy       gen                                                                                31.58

ceval-middle_school_biology                     86817c     accuracy       gen                                                                                71.43

ceval-middle_school_physics                     8accf6     accuracy       gen                                                                                57.89

ceval-middle_school_chemistry                   167a15     accuracy       gen                                                                                80

ceval-veterinary_medicine                       b4e08d     accuracy       gen                                                                                52.17

ceval-college_economics                         f3f4e6     accuracy       gen                                                                                45.45

ceval-business_administration                   c1614e     accuracy       gen                                                                                30.3

ceval-marxism                                   cf874c     accuracy       gen                                                                                47.37

ceval-mao_zedong_thought                        51c7a4     accuracy       gen                                                                                50

ceval-education_science                         591fee     accuracy       gen                                                                                51.72

ceval-teacher_qualification                     4e4ced     accuracy       gen                                                                                72.73

ceval-high_school_politics                      5c0de2     accuracy       gen                                                                                68.42

ceval-high_school_geography                     865461     accuracy       gen                                                                                42.11

ceval-middle_school_politics                    5be3e7     accuracy       gen                                                                                57.14

ceval-middle_school_geography                   8a63be     accuracy       gen                                                                                50

ceval-modern_chinese_history                    fc01af     accuracy       gen                                                                                52.17

ceval-ideological_and_moral_cultivation         a2aa4a     accuracy       gen                                                                                78.95

ceval-logic                                     f5b022     accuracy       gen                                                                                40.91

ceval-law                                       a110a1     accuracy       gen                                                                                33.33

ceval-chinese_language_and_literature           0f8b68     accuracy       gen                                                                                34.78

ceval-art_studies                               2a1300     accuracy       gen                                                                                54.55

ceval-professional_tour_guide                   4e673e     accuracy       gen                                                                                55.17

ceval-legal_professional                        ce8787     accuracy       gen                                                                                30.43

ceval-high_school_chinese                       315705     accuracy       gen                                                                                31.58

ceval-high_school_history                       7eb30a     accuracy       gen                                                                                65

ceval-middle_school_history                     48ab4a     accuracy       gen                                                                                59.09

ceval-civil_servant                             87d061     accuracy       gen                                                                                34.04

ceval-sports_science                            70f27b     accuracy       gen                                                                                63.16

ceval-plant_protection                          8941f9     accuracy       gen                                                                                68.18

ceval-basic_medicine                            c409d6     accuracy       gen                                                                                57.89

ceval-clinical_medicine                         49e82d     accuracy       gen                                                                                54.55

ceval-urban_and_rural_planner                   95b885     accuracy       gen                                                                                52.17

ceval-accountant                                002837     accuracy       gen                                                                                44.9

ceval-fire_engineer                             bc23f5     accuracy       gen                                                                                38.71

ceval-environmental_impact_assessment_engineer  c64e2d     accuracy       gen                                                                                45.16

ceval-tax_accountant                            3a5e3c     accuracy       gen                                                                                34.69

ceval-physician                                 6e277d     accuracy       gen                                                                                57.14

ceval-stem                                      -          naive_average  gen                                                                                46.34

ceval-social-science                            -          naive_average  gen                                                                                51.52

ceval-humanities                                -          naive_average  gen                                                                                48.72

ceval-other                                     -          naive_average  gen                                                                                50.05

ceval-hard                                      -          naive_average  gen                                                                                32.65

ceval                                           -          naive_average  gen                                                                                48.63

config 快速评测，在config 下添加模型配置文件 eval_llama3_8b_demo.py

from mmengine.config import read_base

with read_base():

    from .datasets.mmlu.mmlu_gen_4d595a import mmlu_datasets

datasets = [*mmlu_datasets]

from opencompass.models import HuggingFaceCausalLM

models = [

dict(

type=HuggingFaceCausalLM,

abbr='Llama3_8b', # 运行完结果展示的名称

path='/root/model/Meta-Llama-3-8B-Instruct', # 模型路径

tokenizer_path='/root/model/Meta-Llama-3-8B-Instruct', # 分词器路径

model_kwargs=dict(

device_map='auto',

trust_remote_code=True

),

tokenizer_kwargs=dict(

padding_side='left',

truncation_side='left',

trust_remote_code=True,

use_fast=False

),

generation_kwargs={"eos_token_id": [128001, 128009]},

batch_padding=True,

max_out_len=100,

max_seq_len=2048,

batch_size=16,

run_cfg=dict(num_gpus=1),

)

]

运行python run.py configs/eval_llama3_8b_demo.py

评测完成后，将会看到：

dataset                                            version    metric    mode      Llama3_8b

-------------------------------------------------  ---------  --------  ------  -----------

lukaemon_mmlu_college_biology                      caec7d     accuracy  gen           66.67

lukaemon_mmlu_college_chemistry                    520aa6     accuracy  gen           37

lukaemon_mmlu_college_computer_science             99c216     accuracy  gen           53

lukaemon_mmlu_college_mathematics                  678751     accuracy  gen           36

lukaemon_mmlu_college_physics                      4f382c     accuracy  gen           48.04

lukaemon_mmlu_electrical_engineering               770ce3     accuracy  gen           43.45

lukaemon_mmlu_astronomy                            d3ee01     accuracy  gen           68.42

lukaemon_mmlu_anatomy                              72183b     accuracy  gen           54.07

lukaemon_mmlu_abstract_algebra                     2db373     accuracy  gen           31

lukaemon_mmlu_machine_learning                     0283bb     accuracy  gen           43.75

lukaemon_mmlu_clinical_knowledge                   cb3218     accuracy  gen           58.87

lukaemon_mmlu_global_facts                         ab07b6     accuracy  gen           39

lukaemon_mmlu_management                           80876d     accuracy  gen           78.64

lukaemon_mmlu_nutrition                            4543bd     accuracy  gen           72.55

lukaemon_mmlu_marketing                            7394e3     accuracy  gen           90.17

lukaemon_mmlu_professional_accounting              444b7f     accuracy  gen           49.65

lukaemon_mmlu_high_school_geography                0780e6     accuracy  gen           75.25

lukaemon_mmlu_international_law                    cf3179     accuracy  gen           62.81

lukaemon_mmlu_moral_scenarios                      f6dbe2     accuracy  gen           38.66

lukaemon_mmlu_computer_security                    ce7550     accuracy  gen           35

lukaemon_mmlu_high_school_microeconomics           04d21a     accuracy  gen           62.18

lukaemon_mmlu_professional_law                     5f7e6c     accuracy  gen           47.91

lukaemon_mmlu_medical_genetics                     881ef5     accuracy  gen           62

lukaemon_mmlu_professional_psychology              221a16     accuracy  gen           69.44

lukaemon_mmlu_jurisprudence                        001f24     accuracy  gen           69.44

lukaemon_mmlu_world_religions                      232c09     accuracy  gen           74.85

lukaemon_mmlu_philosophy                           08042b     accuracy  gen           71.06

lukaemon_mmlu_virology                             12e270     accuracy  gen           43.98

lukaemon_mmlu_high_school_chemistry                ae8820     accuracy  gen           42.86

lukaemon_mmlu_public_relations                     e7d39b     accuracy  gen           60

lukaemon_mmlu_high_school_macroeconomics           a01685     accuracy  gen           57.95

lukaemon_mmlu_human_sexuality                      42407c     accuracy  gen           74.05

lukaemon_mmlu_elementary_mathematics               269926     accuracy  gen           28.84

lukaemon_mmlu_high_school_physics                  93278f     accuracy  gen           26.49

lukaemon_mmlu_high_school_computer_science         9965a5     accuracy  gen           63

lukaemon_mmlu_high_school_european_history         eefc90     accuracy  gen           74.55

lukaemon_mmlu_business_ethics                      1dec08     accuracy  gen           51

lukaemon_mmlu_moral_disputes                       a2173e     accuracy  gen           70.81

lukaemon_mmlu_high_school_statistics               8f3f3a     accuracy  gen           52.78

lukaemon_mmlu_miscellaneous                        935647     accuracy  gen           54.15

lukaemon_mmlu_formal_logic                         cfcb0c     accuracy  gen           42.86

lukaemon_mmlu_high_school_government_and_politics  3c52f9     accuracy  gen           86.01

lukaemon_mmlu_prehistory                           bbb197     accuracy  gen           64.2

lukaemon_mmlu_security_studies                     9b1743     accuracy  gen           75.51

lukaemon_mmlu_high_school_biology                  37b125     accuracy  gen           74.84

lukaemon_mmlu_logical_fallacies                    9cebb0     accuracy  gen           68.1

lukaemon_mmlu_high_school_world_history            048e7e     accuracy  gen           83.12

lukaemon_mmlu_professional_medicine                857144     accuracy  gen           72.43

lukaemon_mmlu_high_school_mathematics              ed4dc0     accuracy  gen           31.48

lukaemon_mmlu_college_medicine                     38709e     accuracy  gen           56.65

lukaemon_mmlu_high_school_us_history               8932df     accuracy  gen           82.84

lukaemon_mmlu_sociology                            c266a2     accuracy  gen           76.12

lukaemon_mmlu_econometrics                         d1134d     accuracy  gen           55.26

lukaemon_mmlu_high_school_psychology               7db114     accuracy  gen           65.14

lukaemon_mmlu_human_aging                          82a410     accuracy  gen           62.33

lukaemon_mmlu_us_foreign_policy                    528cfe     accuracy  gen           70

lukaemon_mmlu_conceptual_physics                   63588e     accuracy  gen           26.38

opencompass 官方已经支持 Llama3

https://github.com/open-compass/opencompass/commit/a256753221ad2a33ec9750b31f6284b581c1e1fd#diff-e446451cf0c8fc747c5c720f65f8fa62d7bd7f5c88668692248517d249c798b5

课后总结

1. 非常感谢ModelScope魔搭社区提供的这次机会，可以从一个小白开始一步步学习AI的一些基本知识。开始的时候由于多人创作，一些目录调整导致了一些混乱， 但是很快就进行了修正。现在是一套非常完善，细致，可以手把手教小白入门的课程了。
2. 第一节、第二节是基础课程，相对容易，但是有很多环境配置的工作，很容易出错，涉及的知识面也比较多，这部分大家要耐心仔细。 学习后对于任何AI模型都可以轻松部署了。
3. 第三节有些可惜，免费提供的InterStudio开发机上的显存不够，所以无法进行训练。好在我工作中对于图片的训练需求很少。
4. 第四节通过LMDeploy可以很快部署AI服务， 可以使用两种方式降低显存的使用量，可以通过命令行或API提供服务。
5. 第五节事智能体，虽然跟随课程可以完成训练，但是对智能体的整体知识的介绍在课程中是比较少的，有机会还要自己学习。
6. 第六节由于下载速度的问题没有完成，希望OpenCompass首页_上海人工智能实验室 (shlab.org.cn)可以对国内的用户可以提供更多的带宽。
7. 这次学习整体上是小白入门教程，仅仅打开了AI的一个窗口，后续需要扩展的知识还有很多，希望官方可以提供一个成长路径，为后续学习提供指导。