使用Lora技术进行Dreambooth训练【抢先体验版】

最新推荐文章于 2024-05-27 22:16:09 发布

AI Studio

最新推荐文章于 2024-05-27 22:16:09 发布

阅读量6.1k

点赞数

文章标签： python 深度学习人工智能

原文链接：https://aistudio.baidu.com/aistudio/projectdetail/5481677?channelType=0&channel=0

版权

#★★★本文源自AlStudio社区精品项目，
【点击此处】查看更多精品内容 >>>
(https://aistudio.baidu.com/aistudio/proiectoverview/public?ad-from=4100)

使用Lora技术进行Dreambooth训练【抢先体验版】

LoRA: Low-Rank Adaptation of Large Language Models 是微软研究员引入的一项新技术，主要用于处理大模型微调的问题。目前超过数十亿以上参数的具有强能力的大模型 (例如 GPT-3) 通常在为了适应其下游任务的微调中会呈现出巨大开销。LoRA 建议冻结预训练模型的权重并在每个 Transformer 块中注入可训练层 (秩-分解矩阵)。因为不需要为大多数模型权重计算梯度，所以大大减少了需要训练参数的数量并且降低了 GPU 的内存要求。研究人员发现，通过聚焦大模型的 Transformer 注意力块，使用 LoRA 进行的微调质量与全模型微调相当，同时速度更快且需要更少的计算。

论文链接：https://arxiv.org/abs/2106.09685

参考代码：https://github.com/huggingface/diffusers/tree/main/examples/dreambooth

中文介绍：https://mp.weixin.qq.com/s/kEGwA_7qAKhIuoxPJyfNuw

1. 安装依赖

运行下面的按钮安装依赖，为了确保安装成功，安装完毕请重启内核！（注意：这里只需要运行一次！）

!pip install -U paddlenlp ppdiffusers safetensors --user

Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Requirement already satisfied: paddlenlp in ./.data/webide/pip/lib/python3.7/site-packages (2.5.1)
Requirement already satisfied: ppdiffusers in ./.data/webide/pip/lib/python3.7/site-packages (0.11.0)
Requirement already satisfied: safetensors in ./.data/webide/pip/lib/python3.7/site-packages (0.2.8)
Requirement already satisfied: paddlefsl in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (1.1.0)
Requirement already satisfied: sentencepiece in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (0.1.96)
Requirement already satisfied: huggingface-hub>=0.11.1 in ./.data/webide/pip/lib/python3.7/site-packages (from paddlenlp) (0.12.0)
Requirement already satisfied: seqeval in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (1.2.2)
Requirement already satisfied: fastapi in ./.data/webide/pip/lib/python3.7/site-packages (from paddlenlp) (0.91.0)
Requirement already satisfied: typer in ./.data/webide/pip/lib/python3.7/site-packages (from paddlenlp) (0.7.0)
Requirement already satisfied: multiprocess<=0.70.12.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (0.70.11.1)
Requirement already satisfied: jieba in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (0.42.1)
Requirement already satisfied: Flask-Babel<3.0.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (1.0.0)
Requirement already satisfied: rich in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (12.6.0)
Requirement already satisfied: datasets>=2.0.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (2.7.0)
Requirement already satisfied: paddle2onnx in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (1.0.0)
Requirement already satisfied: colorama in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (0.4.4)
Requirement already satisfied: visualdl in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (2.4.0)
Requirement already satisfied: uvicorn in ./.data/webide/pip/lib/python3.7/site-packages (from paddlenlp) (0.20.0)
Requirement already satisfied: tqdm in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (4.64.1)
Requirement already satisfied: dill<0.3.5 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (0.3.3)
Requirement already satisfied: colorlog in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from paddlenlp) (4.1.0)
Requirement already satisfied: Pillow in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from ppdiffusers) (8.2.0)
Requirement already satisfied: regex in ./.data/webide/pip/lib/python3.7/site-packages (from ppdiffusers) (2022.10.31)
Requirement already satisfied: ftfy in ./.data/webide/pip/lib/python3.7/site-packages (from ppdiffusers) (6.1.1)
Requirement already satisfied: pyyaml>=5.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (5.1.2)
Requirement already satisfied: numpy>=1.17 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (1.19.5)
Requirement already satisfied: pyarrow>=6.0.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (10.0.0)
Requirement already satisfied: fsspec[http]>=2021.11.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (2022.11.0)
Requirement already satisfied: requests>=2.19.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (2.24.0)
Requirement already satisfied: xxhash in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (3.1.0)
Requirement already satisfied: importlib-metadata in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (4.2.0)
Requirement already satisfied: packaging in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (21.3)
Requirement already satisfied: aiohttp in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (3.8.3)
Requirement already satisfied: responses<0.19 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (0.18.0)
Requirement already satisfied: pandas in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from datasets>=2.0.0->paddlenlp) (1.1.5)
Requirement already satisfied: Jinja2>=2.5 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask-Babel<3.0.0->paddlenlp) (3.0.0)
Requirement already satisfied: pytz in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask-Babel<3.0.0->paddlenlp) (2019.3)
Requirement already satisfied: Babel>=2.3 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask-Babel<3.0.0->paddlenlp) (2.8.0)
Requirement already satisfied: Flask in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask-Babel<3.0.0->paddlenlp) (1.1.1)
Requirement already satisfied: filelock in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from huggingface-hub>=0.11.1->paddlenlp) (3.0.12)
Requirement already satisfied: typing-extensions>=3.7.4.3 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from huggingface-hub>=0.11.1->paddlenlp) (4.3.0)
Requirement already satisfied: starlette<0.25.0,>=0.24.0 in ./.data/webide/pip/lib/python3.7/site-packages (from fastapi->paddlenlp) (0.24.0)
Requirement already satisfied: pydantic!=1.7,!=1.7.1,!=1.7.2,!=1.7.3,!=1.8,!=1.8.1,<2.0.0,>=1.6.2 in ./.data/webide/pip/lib/python3.7/site-packages (from fastapi->paddlenlp) (1.10.4)
Requirement already satisfied: wcwidth>=0.2.5 in ./.data/webide/pip/lib/python3.7/site-packages (from ftfy->ppdiffusers) (0.2.6)
Requirement already satisfied: pygments<3.0.0,>=2.6.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from rich->paddlenlp) (2.13.0)
Requirement already satisfied: commonmark<0.10.0,>=0.9.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from rich->paddlenlp) (0.9.1)
Requirement already satisfied: scikit-learn>=0.21.3 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from seqeval->paddlenlp) (0.24.2)
Requirement already satisfied: click<9.0.0,>=7.1.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from typer->paddlenlp) (8.0.4)
Requirement already satisfied: h11>=0.8 in ./.data/webide/pip/lib/python3.7/site-packages (from uvicorn->paddlenlp) (0.14.0)
Requirement already satisfied: matplotlib in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->paddlenlp) (2.2.3)
Requirement already satisfied: six>=1.14.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->paddlenlp) (1.16.0)
Requirement already satisfied: protobuf>=3.11.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->paddlenlp) (3.20.0)
Requirement already satisfied: bce-python-sdk in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from visualdl->paddlenlp) (0.8.53)
Requirement already satisfied: Werkzeug>=0.15 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask->Flask-Babel<3.0.0->paddlenlp) (0.16.0)
Requirement already satisfied: itsdangerous>=0.24 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Flask->Flask-Babel<3.0.0->paddlenlp) (1.1.0)
Requirement already satisfied: attrs>=17.3.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (22.1.0)
Requirement already satisfied: aiosignal>=1.1.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (1.2.0)
Requirement already satisfied: multidict<7.0,>=4.5 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (6.0.2)
Requirement already satisfied: async-timeout<5.0,>=4.0.0a3 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (4.0.2)
Requirement already satisfied: asynctest==0.13.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (0.13.0)
Requirement already satisfied: frozenlist>=1.1.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (1.3.0)
Requirement already satisfied: yarl<2.0,>=1.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (1.7.2)
Requirement already satisfied: charset-normalizer<3.0,>=2.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from aiohttp->datasets>=2.0.0->paddlenlp) (2.1.1)
Requirement already satisfied: MarkupSafe>=2.0.0rc2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from Jinja2>=2.5->Flask-Babel<3.0.0->paddlenlp) (2.0.1)
Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from packaging->datasets>=2.0.0->paddlenlp) (3.0.9)
Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests>=2.19.0->datasets>=2.0.0->paddlenlp) (2019.9.11)
Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests>=2.19.0->datasets>=2.0.0->paddlenlp) (1.25.11)
Requirement already satisfied: chardet<4,>=3.0.2 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests>=2.19.0->datasets>=2.0.0->paddlenlp) (3.0.4)
Requirement already satisfied: idna<3,>=2.5 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from requests>=2.19.0->datasets>=2.0.0->paddlenlp) (2.8)
Requirement already satisfied: scipy>=0.19.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from scikit-learn>=0.21.3->seqeval->paddlenlp) (1.6.3)
Requirement already satisfied: joblib>=0.11 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from scikit-learn>=0.21.3->seqeval->paddlenlp) (0.14.1)
Requirement already satisfied: threadpoolctl>=2.0.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from scikit-learn>=0.21.3->seqeval->paddlenlp) (2.1.0)
Requirement already satisfied: anyio<5,>=3.4.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from starlette<0.25.0,>=0.24.0->fastapi->paddlenlp) (3.6.1)
Requirement already satisfied: future>=0.6.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from bce-python-sdk->visualdl->paddlenlp) (0.18.0)
Requirement already satisfied: pycryptodome>=3.8.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from bce-python-sdk->visualdl->paddlenlp) (3.9.9)
Requirement already satisfied: zipp>=0.5 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from importlib-metadata->datasets>=2.0.0->paddlenlp) (3.8.1)
Requirement already satisfied: kiwisolver>=1.0.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from matplotlib->visualdl->paddlenlp) (1.1.0)
Requirement already satisfied: python-dateutil>=2.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from matplotlib->visualdl->paddlenlp) (2.8.2)
Requirement already satisfied: cycler>=0.10 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from matplotlib->visualdl->paddlenlp) (0.10.0)
Requirement already satisfied: sniffio>=1.1 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from anyio<5,>=3.4.0->starlette<0.25.0,>=0.24.0->fastapi->paddlenlp) (1.3.0)
Requirement already satisfied: setuptools in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from kiwisolver>=1.0.1->matplotlib->visualdl->paddlenlp) (56.2.0)

[1m[[0m[34;49mnotice[0m[1;39;49m][0m[39;49m A new release of pip available: [0m[31;49m22.1.2[0m[39;49m -> [0m[32;49m23.0.1[0m
[1m[[0m[34;49mnotice[0m[1;39;49m][0m[39;49m To update, run: [0m[32;49mpip install --upgrade pip[0m

2. 准备要训练的图片

在这里我们已经在dogs文件夹准备好了如下所示的5张图片。

3. 开始训练

参数解释：

pretrained_model_name_or_path ：想要训练的模型名称，例如：“runwayml/stable-diffusion-v1-5”，更多模型可参考 paddlenlp 文档
instance_data_dir：想要训练的图片地址。
instance_prompt：训练的prompt文本。
resolution：训练时图像的大小，建议为512。
train_batch_size：训练时候使用的batch_size，可不修改。
gradient_accumulation_steps：梯度累积的步数，可不修改。
checkpointing_steps：每隔多少步保存模型。
learning_rate：训练使用的学习率。
report_to：我们将训练过程中出的图片导出到visudl工具中。
lr_scheduler：学习率衰减策略，可以是：“linear”, “constant”, “cosine”，"cosine_with_restarts"等。
lr_warmup_steps：学习率衰减前，warmup到最大学习率所需要的步数。
max_train_steps：最多训练多少步。
validation_prompt：训练的过程中我们会评估训练的怎么样，因此我们需要设置评估使用的prompt文本。
validation_epochs：每隔多少个epoch评估模型，我们可以查看训练的进度条，知道当前到了第几个epoch。
validation_guidance_scale：评估过程中的CFG引导值，默认为5.0.
seed：随机种子，设置后可以复现训练结果。
lora_rank：lora 的 rank值，默认为128，与开源的版本保持一致。
use_lion：表示是否使用lion优化器，如果我们不想使用lion的话需要把 --use_lion True 表示使用 --use_lion False 表示不使用。
lora_weight_or_path：我们需要加载的lora权重，当前支持：pt，ckpt，safetensors，和pdparams这些格式，可直接加载这里的lora权重 https://civitai.com/models。

注意：

会保存2种格式的权重，一个是paddle的，一个是safetensors的，可以使用 https://github.com/bmaltais/kohya_ss 这个人的加载。

dreambooth lora

!python train_dreambooth_lora.py \
  --pretrained_model_name_or_path="runwayml/stable-diffusion-v1-5"  \
  --instance_data_dir="./dogs" \
  --output_dir="./dream_booth_lora_outputs" \
  --instance_prompt="a photo of sks dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --checkpointing_steps=100 \
  --learning_rate=1e-4 \
  --report_to="visualdl" \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=500 \
  --lora_rank=128 \
  --validation_prompt="A photo of sks dog in a bucket" \
  --validation_epochs=25 \
  --validation_guidance_scale=5.0 \
  --use_lion False \
  --seed=0

[33m[2023-02-23 10:13:09,015] [ WARNING][0m - You are using a model of type clip_text_model to instantiate a model of type . This is not supported for all configurations of models and can yield errors.[0m
W0223 10:13:09.018703 11490 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 11.2
W0223 10:13:09.022612 11490 gpu_resources.cc:91] device: 0, cuDNN Version: 8.2.
Train Steps:  20%|███████████████████████▌                                                                                              | 100/500 [01:12<03:04,  2.16it/s, epoch=0019, step_loss=0.0413]Saved lora weights to ./dream_booth_lora_outputs/checkpoint-100
Train Steps:  40%|███████████████████████████████████████████████▌                                                                       | 200/500 [02:25<02:19,  2.15it/s, epoch=0039, step_loss=0.446]Saved lora weights to ./dream_booth_lora_outputs/checkpoint-200
Train Steps:  60%|██████████████████████████████████████████████████████████████████████▏                                              | 300/500 [03:37<01:33,  2.13it/s, epoch=0059, step_loss=0.00273]Saved lora weights to ./dream_booth_lora_outputs/checkpoint-300
Train Steps:  80%|███████████████████████████████████████████████████████████████████████████████████████████████▏                       | 400/500 [04:53<00:47,  2.11it/s, epoch=0079, step_loss=0.275]Saved lora weights to ./dream_booth_lora_outputs/checkpoint-400
Train Steps: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 500/500 [05:44<00:00,  2.14it/s, epoch=0099, step_loss=0.00985]Saved lora weights to ./dream_booth_lora_outputs/checkpoint-500
Saved final lora weights to ./dream_booth_lora_outputs
Train Steps: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 500/500 [05:48<00:00,  1.43it/s, epoch=0099, step_loss=0.00985]
[0m

文生图 lora

–train_data_dir 这个需要放图文对的文件夹，里面是图片和txt。

–image_format 表示 train_data_dir 文件夹内的图片格式，比如png，jpg，jpeg

–use_lion Fasle 表示不使用lion优化器。

!python train_text_to_image_lora.py \
  --pretrained_model_name_or_path="runwayml/stable-diffusion-v1-5"  \
  --output_dir="./text_to_image_lora_outputs3" \
  --train_data_dir="mishanwu" \
  --image_format="png" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --checkpointing_steps=500 \
  --learning_rate=6e-5 \
  --report_to="visualdl" \
  --lr_scheduler="cosine_with_restarts" \
  --lr_warmup_steps=0 \
  --max_train_steps=1000 \
  --lora_rank=128 \
  --validation_prompt="1girl, solo, black_background, looking_at_viewer, parted_lips, tears, brown_eyes" \
  --validation_epochs=1 \
  --validation_guidance_scale=5.0 \
  --use_lion False \
  --seed=0

[33m[2023-02-23 11:11:34,386] [ WARNING][0m - You are using a model of type clip_text_model to instantiate a model of type . This is not supported for all configurations of models and can yield errors.[0m
W0223 11:11:34.389356  7327 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 11.2
W0223 11:11:34.392953  7327 gpu_resources.cc:91] device: 0, cuDNN Version: 8.2.
Resolving data files: 100%|████████████████| 581/581 [00:00<00:00, 50722.06it/s]
Using custom data configuration default-4fb24e511b7f6118
Downloading and preparing dataset imagefolder/default to /home/aistudio/.cache/huggingface/datasets/imagefolder/default-4fb24e511b7f6118/0.0.0/37fbb85cc714a338bea574ac6c7d0b5be5aff46c1862c1989b20e0771199e93f...
Downloading data files #0:   0%|                        | 0/19 [00:00<?, ?obj/s]


Downloading data files #3:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A
Downloading data files #1:   0%|                        | 0/19 [00:00<?, ?obj/s][A






Downloading data files #7:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A





Downloading data files #6:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A



Downloading data files #4:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A

Downloading data files #0: 100%|█████████████| 19/19 [00:00<00:00, 3045.97obj/s][A[A
Downloading data files #3: 100%|█████████████| 18/18 [00:00<00:00, 2919.24obj/s]










Downloading data files #10:   0%|                       | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A










Downloading data files #11:   0%|                       | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A




Downloading data files #1: 100%|█████████████| 19/19 [00:00<00:00, 2589.92obj/s][A[A[A[A[A
Downloading data files #4: 100%|█████████████| 18/18 [00:00<00:00, 4414.54obj/s]
Downloading data files #7: 100%|█████████████| 18/18 [00:00<00:00, 2468.45obj/s]









Downloading data files #6: 100%|█████████████| 18/18 [00:00<00:00, 2742.57obj/s][A[A[A[A[A[A[A[A[A















Downloading data files #2: 100%|█████████████| 19/19 [00:00<00:00, 3250.74obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A[A[A












Downloading data files #12:   0%|                       | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A[A












Downloading data files #13:   0%|                       | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A







Downloading data files #8:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A













Downloading data files #10: 100%|████████████| 18/18 [00:00<00:00, 2681.49obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A[A
Downloading data files #11: 100%|████████████| 18/18 [00:00<00:00, 2829.21obj/s]
Downloading data files #15: 100%|████████████| 18/18 [00:00<00:00, 5331.37obj/s]
Downloading data files #5: 100%|█████████████| 18/18 [00:00<00:00, 2877.96obj/s]
Downloading data files #13: 100%|████████████| 18/18 [00:00<00:00, 7175.88obj/s]
Downloading data files #14: 100%|████████████| 18/18 [00:00<00:00, 9971.93obj/s]
Downloading data files #8: 100%|█████████████| 18/18 [00:00<00:00, 6747.47obj/s]
Downloading data files #12: 100%|████████████| 18/18 [00:00<00:00, 4458.07obj/s]
Downloading data files #9: 100%|█████████████| 18/18 [00:00<00:00, 2966.27obj/s]


Downloading data files #2:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A


Downloading data files #3:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A



Downloading data files #2: 100%|█████████████| 18/18 [00:00<00:00, 8158.36obj/s][A[A[A[A
Downloading data files #0:   0%|                        | 0/19 [00:00<?, ?obj/s]




Downloading data files #4: 100%|█████████████| 18/18 [00:00<00:00, 7530.17obj/s][A[A[A[A[A
Downloading data files #3: 100%|█████████████| 18/18 [00:00<00:00, 5302.16obj/s]
Downloading data files #0: 100%|█████████████| 19/19 [00:00<00:00, 8757.34obj/s]








Downloading data files #8:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A








Downloading data files #9:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A






Downloading data files #7:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A
Downloading data files #1:   0%|                        | 0/19 [00:00<?, ?obj/s][A










Downloading data files #5: 100%|█████████████| 18/18 [00:00<00:00, 4015.18obj/s][A[A[A[A[A[A[A[A[A[A[A






Downloading data files #6:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A









Downloading data files #8: 100%|█████████████| 18/18 [00:00<00:00, 4227.42obj/s][A[A[A[A[A[A[A[A[A[A
Downloading data files #7: 100%|█████████████| 18/18 [00:00<00:00, 4900.21obj/s]
Downloading data files #6: 100%|█████████████| 18/18 [00:00<00:00, 9804.87obj/s]












Downloading data files #11: 100%|████████████| 18/18 [00:00<00:00, 4705.36obj/s][A[A[A[A[A[A[A[A[A[A[A[A
Downloading data files #9: 100%|█████████████| 18/18 [00:00<00:00, 3152.95obj/s]
Downloading data files #10: 100%|████████████| 18/18 [00:00<00:00, 6718.65obj/s]
Downloading data files #1: 100%|█████████████| 19/19 [00:00<00:00, 3689.60obj/s]















Downloading data files #15:   0%|                       | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A[A[A












Downloading data files #13:   0%|                       | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A













Downloading data files #12: 100%|████████████| 18/18 [00:00<00:00, 6109.20obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A[A
Downloading data files #15: 100%|███████████| 18/18 [00:00<00:00, 11363.26obj/s]
Downloading data files #13: 100%|███████████| 18/18 [00:00<00:00, 13148.29obj/s]
Downloading data files #14: 100%|███████████| 18/18 [00:00<00:00, 16183.81obj/s]
Extracting data files #0:   0%|                         | 0/19 [00:00<?, ?obj/s]


Extracting data files #3:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A[A

Extracting data files #2:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A
Extracting data files #1:   0%|                         | 0/19 [00:00<?, ?obj/s][A






Extracting data files #7:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A








Extracting data files #9:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A




Extracting data files #5:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A[A[A[A





Extracting data files #6:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A









Extracting data files #10:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A







Extracting data files #8:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A



Extracting data files #4:   0%|                         | 0/18 [00:00<?, ?obj/s][A[A[A[A










Extracting data files #11:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A











Extracting data files #12:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A[A












Extracting data files #13:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A














Extracting data files #15:   0%|                        | 0/18 [00:00<?, ?obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A[A[A













Extracting data files #3: 100%|██████████████| 18/18 [00:00<00:00, 1565.14obj/s][A[A[A[A[A[A[A[A[A[A[A[A[A[A
Extracting data files #0: 100%|██████████████| 19/19 [00:00<00:00, 1405.75obj/s]
Extracting data files #2: 100%|██████████████| 18/18 [00:00<00:00, 1560.64obj/s]
Extracting data files #1: 100%|██████████████| 19/19 [00:00<00:00, 1753.47obj/s]
Extracting data files #7: 100%|██████████████| 18/18 [00:00<00:00, 1907.66obj/s]
Extracting data files #9: 100%|██████████████| 18/18 [00:00<00:00, 1972.81obj/s]
Extracting data files #5: 100%|██████████████| 18/18 [00:00<00:00, 2031.58obj/s]
Extracting data files #6: 100%|██████████████| 18/18 [00:00<00:00, 2102.00obj/s]
Extracting data files #8: 100%|██████████████| 18/18 [00:00<00:00, 2238.62obj/s]
Extracting data files #12: 100%|█████████████| 18/18 [00:00<00:00, 2500.58obj/s]
Extracting data files #10: 100%|█████████████| 18/18 [00:00<00:00, 1698.37obj/s]
Extracting data files #13: 100%|█████████████| 18/18 [00:00<00:00, 2295.87obj/s]
Extracting data files #4: 100%|██████████████| 18/18 [00:00<00:00, 1713.79obj/s]
Extracting data files #11: 100%|█████████████| 18/18 [00:00<00:00, 1759.56obj/s]
Extracting data files #15: 100%|█████████████| 18/18 [00:00<00:00, 2079.59obj/s]
Extracting data files #14: 100%|█████████████| 18/18 [00:00<00:00, 1828.07obj/s]
Dataset imagefolder downloaded and prepared to /home/aistudio/.cache/huggingface/datasets/imagefolder/default-4fb24e511b7f6118/0.0.0/37fbb85cc714a338bea574ac6c7d0b5be5aff46c1862c1989b20e0771199e93f. Subsequent calls will reuse this data.
100%|████████████████████████████████████████████| 1/1 [00:00<00:00, 256.14it/s]
Train Steps:  29%|▎| 290/1000 [02:14<05:22,  2.20it/s, epoch=0000, step_loss=0.1You have disabled the safety checker for <class 'ppdiffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline'> by passing `safety_checker=None`. Ensure that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered results in services or applications open to the public. PaddleNLP team, diffusers team and Hugging Face strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling it only for use-cases that involve analyzing network behavior or auditing its results. For more information, please have a look at https://github.com/huggingface/diffusers/pull/254 .
Train Steps:  50%|▌| 500/1000 [04:16<03:54,  2.13it/s, epoch=0001, step_loss=0.0Saved lora weights to ./text_to_image_lora_outputs3/checkpoint-500
Train Steps:  58%|▌| 580/1000 [04:55<03:09,  2.22it/s, epoch=0001, step_loss=0.2You have disabled the safety checker for <class 'ppdiffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline'> by passing `safety_checker=None`. Ensure that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered results in services or applications open to the public. PaddleNLP team, diffusers team and Hugging Face strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling it only for use-cases that involve analyzing network behavior or auditing its results. For more information, please have a look at https://github.com/huggingface/diffusers/pull/254 .
Train Steps:  87%|▊| 870/1000 [07:40<00:58,  2.22it/s, epoch=0002, step_loss=0.0You have disabled the safety checker for <class 'ppdiffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline'> by passing `safety_checker=None`. Ensure that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered results in services or applications open to the public. PaddleNLP team, diffusers team and Hugging Face strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling it only for use-cases that involve analyzing network behavior or auditing its results. For more information, please have a look at https://github.com/huggingface/diffusers/pull/254 .
Train Steps: 100%|█| 1000/1000 [09:10<00:00,  1.25it/s, epoch=0003, step_loss=0.Saved lora weights to ./text_to_image_lora_outputs3/checkpoint-1000
You have disabled the safety checker for <class 'ppdiffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline'> by passing `safety_checker=None`. Ensure that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered results in services or applications open to the public. PaddleNLP team, diffusers team and Hugging Face strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling it only for use-cases that involve analyzing network behavior or auditing its results. For more information, please have a look at https://github.com/huggingface/diffusers/pull/254 .
Saved final lora weights to ./text_to_image_lora_outputs3
Train Steps: 100%|█| 1000/1000 [09:44<00:00,  1.71it/s, epoch=0003, step_loss=0.
[0m

4. 启动visualdl程序，查看我们训练过程中出图情况

5. 加载训练好的文件进行推理

import lora_helper
from allinone import StableDiffusionPipelineAllinOne
from ppdiffusers import DPMSolverMultistepScheduler
import paddle
# 基础模型，需要是paddle版本的权重，未来会加更多的权重
pretrained_model_name_or_path = "runwayml/stable-diffusion-v1-5"

# 我们加载safetensor版本的权重
lora_outputs_path = "9070.safetensors"

# 加载之前的模型
pipe = StableDiffusionPipelineAllinOne.from_pretrained(pretrained_model_name_or_path, safety_checker=None)
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)

# 加载lora权重
from IPython.display import clear_output, display
clear_output()
pipe.apply_lora(lora_outputs_path)

|---------------当前的rank是 128！
|---------------当前的alpha是 128.0！
Loading lora_weights successfully!

“runwayml/stable-diffusion-v1-5”

我们加载safetensor版本的权重

lora_outputs_path = “9070.safetensors”

加载之前的模型

pipe = StableDiffusionPipelineAllinOne.from_pretrained(pretrained_model_name_or_path, safety_checker=None)
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)

加载lora权重

from IPython.display import clear_output, display
clear_output()
pipe.apply_lora(lora_outputs_path)


    |---------------当前的rank是 128！
    |---------------当前的alpha是 128.0！
    Loading lora_weights successfully!



```python
import lora_helper
from allinone import StableDiffusionPipelineAllinOne
from ppdiffusers import DPMSolverMultistepScheduler

prompt               = "A photo of sks dog in a bucket"
negative_prompt      = ""
guidance_scale       = 8
num_inference_steps  = 25
height               = 512
width                = 512

img = pipe(prompt, negative_prompt=negative_prompt, guidance_scale=guidance_scale, height=height, width=width, num_inference_steps=num_inference_steps).images[0]
display(img)
display(img.argument)

  0%|          | 0/25 [00:00<?, ?it/s]

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-3HTTZjvS-1677160871054)(main_files/main_12_1.png)]

{'prompt': 'A photo of sks dog in a bucket',
 'negative_prompt': '',
 'height': 512,
 'width': 512,
 'num_inference_steps': 25,
 'guidance_scale': 8,
 'num_images_per_prompt': 1,
 'eta': 0.0,
 'seed': 3574959348,
 'latents': None,
 'max_embeddings_multiples': 1,
 'no_boseos_middle': False,
 'skip_parsing': False,
 'skip_weighting': False,
 'epoch_time': 1676862593.5281246}

AI Studio

关注

0
点赞
踩
10

收藏

觉得还不错? 一键收藏
1
评论
使用Lora技术进行Dreambooth训练【抢先体验版】

LoRA: Low-Rank Adaptation of Large Language Models 是微软研究员引入的一项新技术，主要用于处理大模型微调的问题。目前超过数十亿以上参数的具有强能力的大模型 (例如 GPT-3) 通常在为了适应其下游任务的微调中会呈现出巨大开销。LoRA 建议冻结预训练模型的权重并在每个 Transformer 块中注入可训练层 (秩-分解矩阵)。因为不需要为大多数模型权重计算梯度，所以大大减少了需要训练参数的数量并且降低了 GPU 的内存要求。研究人员发现，通过聚焦大模型的
复制链接

扫一扫