jochemstoel · February 1, 2023 23:22
diff --git a/dreambooth_stable_diffusion_mod.ipynb b/dreambooth_stable_diffusion_mod.ipynb
 {
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/thx-pw/37ba770dfd2b15f119ec8032c3ae90a1/dreambooth_stable_diffusion_mod.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "XU7NuMAA2drw"
      },
      "outputs": [],
      "source": [
        "#@markdown Check type of GPU and VRAM available.\n",
        "!nvidia-smi --query-gpu=name,memory.total,memory.free --format=csv,noheader"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BzM7j0ZSc_9c"
      },
      "source": [
        "https://github.com/ShivamShrirao/diffusers/tree/main/examples/dreambooth"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wnTMyW41cC1E"
      },
      "source": [
        "## Install Requirements"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "aLWXPZqjsZVV"
      },
      "outputs": [],
      "source": [
        "!wget -q https://github.com/ShivamShrirao/diffusers/raw/main/examples/dreambooth/train_dreambooth.py\n",
        "!wget -q https://github.com/ShivamShrirao/diffusers/raw/main/scripts/convert_diffusers_to_original_stable_diffusion.py\n",
        "%pip install -qq git+https://github.com/ShivamShrirao/diffusers\n",
        "%pip install -q -U --pre triton\n",
        "%pip install -q accelerate transformers ftfy bitsandbytes==0.35.0 gradio natsort"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "y4lqqWT_uxD2"
      },
      "outputs": [],
      "source": [
        "#@title Login to HuggingFace 🤗\n",
        "\n",
        "#@markdown You need to accept the model license before downloading or using the Stable Diffusion weights. Please, visit the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5), read the license and tick the checkbox if you agree. You have to be a registered user in 🤗 Hugging Face Hub, and you'll also need to use an access token for the code to work.\n",
        "# https://huggingface.co/settings/tokens\n",
        "!mkdir -p ~/.huggingface\n",
        "HUGGINGFACE_TOKEN = \"\" #@param {type:\"string\"}\n",
        "!echo -n \"{HUGGINGFACE_TOKEN}\" > ~/.huggingface/token"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XfTlc8Mqb8iH"
      },
      "source": [
        "### Install xformers from precompiled wheel."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "n6dcjPnnaiCn"
      },
      "outputs": [],
      "source": [
        "%pip install --no-deps -q https://github.com/brian6091/xformers-wheels/releases/download/0.0.15.dev0%2B4c06c79/xformers-0.0.15.dev0+4c06c79.d20221205-cp38-cp38-linux_x86_64.whl\n",
        "# These were compiled on Tesla T4.\n",
        "\n",
        "# If precompiled wheels don't work, install it with the following command. It will take around 40 minutes to compile.\n",
        "# %pip install git+https://github.com/facebookresearch/xformers@4c06c79#egg=xformers"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "G0NV324ZcL9L"
      },
      "source": [
        "## Settings and run"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Rxg0y5MBudmd",
        "cellView": "form"
      },
      "outputs": [],
      "source": [
        "#@markdown Name/Path of the initial model.\n",
        "MODEL_NAME = \"hakurei/waifu-diffusion\" #@param [\"hakurei/waifu-diffusion\", \"runwayml/stable-diffusion-v1-5\", \"CompVis/stable-diffusion-v1-4\", \"naclbit/trinart_stable_diffusion_v2,diffusers-115k\", \"naclbit/trinart_stable_diffusion_v2,diffusers-95k\", \"naclbit/trinart_stable_diffusion_v2,diffusers-60k\"] {allow-input: true}\n",
        "\n",
        "INSTANCE_NAME = \"sks\" #@param {type:\"string\"}\n",
        "CLASS_NAME = \"1boy\" #@param {type:\"string\"}\n",
        "INSTANCE_PROMPT = f\"{INSTANCE_NAME} {CLASS_NAME}\"\n",
        "\n",
        "CLASS_DIR = f\"/content/data/{CLASS_NAME}\"\n",
        "\n",
        "INSTANCE_DIR = f\"/content/data/{INSTANCE_NAME}\"\n",
        "!mkdir -p $INSTANCE_DIR\n",
        "\n",
        "OUTPUT_DIR = f\"/content/stable_diffusion_weights/{INSTANCE_NAME}\"\n",
        "print(f\"[*] Weights will be saved at {OUTPUT_DIR}\")\n",
        "!mkdir -p $OUTPUT_DIR"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "fe-GgtnUVO_e"
      },
      "outputs": [],
      "source": [
        "#@markdown Upload your images by running this cell.\n",
        "\n",
        "#@markdown OR\n",
        "\n",
        "#@markdown You can use the file manager on the left panel to upload (drag and drop) to INSTANCE_DIR (it uploads faster)\n",
        "\n",
        "import os\n",
        "from google.colab import files\n",
        "import shutil\n",
        "\n",
        "uploaded = files.upload()\n",
        "for filename in uploaded.keys():\n",
        "    dst_path = os.path.join(INSTANCE_DIR, filename)\n",
        "    shutil.move(filename, dst_path)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qn5ILIyDJIcX"
      },
      "source": [
        "# Start Training\n",
        "\n",
        "Use the table below to choose the best flags based on your memory and speed requirements. Tested on Tesla T4 GPU.\n",
        "\n",
        "\n",
        "| `fp16` | `train_batch_size` | `gradient_accumulation_steps` | `gradient_checkpointing` | `use_8bit_adam` | GB VRAM usage | Speed (it/s) |\n",
        "| ---- | ------------------ | ----------------------------- | ----------------------- | --------------- | ---------- | ------------ |\n",
        "| fp16 | 1                  | 1                             | TRUE                    | TRUE            | 9.92       | 0.93         |\n",
        "| no   | 1                  | 1                             | TRUE                    | TRUE            | 10.08      | 0.42         |\n",
        "| fp16 | 2                  | 1                             | TRUE                    | TRUE            | 10.4       | 0.66         |\n",
        "| fp16 | 1                  | 1                             | FALSE                   | TRUE            | 11.17      | 1.14         |\n",
        "| no   | 1                  | 1                             | FALSE                   | TRUE            | 11.17      | 0.49         |\n",
        "| fp16 | 1                  | 2                             | TRUE                    | TRUE            | 11.56      | 1            |\n",
        "| fp16 | 2                  | 1                             | FALSE                   | TRUE            | 13.67      | 0.82         |\n",
        "| fp16 | 1                  | 2                             | FALSE                   | TRUE            | 13.7       | 0.83          |\n",
        "| fp16 | 1                  | 1                             | TRUE                    | FALSE           | 15.79      | 0.77         |\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-ioxxvHoicPs"
      },
      "source": [
        "Add `--gradient_checkpointing` flag for around 9.92 GB VRAM usage.\n",
        "\n",
        "remove `--use_8bit_adam` flag for full precision. Requires 15.79 GB with `--gradient_checkpointing` else 17.8 GB."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "jjcSXTp-u-Eg"
      },
      "outputs": [],
      "source": [
        "!accelerate launch train_dreambooth.py \\\n",
        "  --pretrained_model_name_or_path=$MODEL_NAME \\\n",
        "  --pretrained_vae_name_or_path=\"\" \\\n",
        "  --instance_data_dir=$INSTANCE_DIR \\\n",
        "  --class_data_dir=$CLASS_DIR \\\n",
        "  --output_dir=$OUTPUT_DIR \\\n",
        "  --revision=\"fp16\" \\\n",
        "  --with_prior_preservation --prior_loss_weight=1.0 \\\n",
        "  --resolution=512 \\\n",
        "  --train_batch_size=4 \\\n",
        "  --train_text_encoder \\\n",
        "  --pad_tokens \\\n",
        "  --mixed_precision=\"fp16\" \\\n",
        "  --gradient_checkpointing \\\n",
        "  --use_8bit_adam \\\n",
        "  --gradient_accumulation_steps=1 \\\n",
        "  --learning_rate=1e-6 \\\n",
        "  --lr_scheduler=\"constant\" \\\n",
        "  --lr_warmup_steps=0 \\\n",
        "  --num_class_images=60 \\\n",
        "  --sample_batch_size=1 \\\n",
        "  --max_train_steps=300 \\\n",
        "  --instance_prompt=\"{INSTANCE_PROMPT}\" \\\n",
        "  --class_prompt=\"{CLASS_NAME}\""
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5V8wgU0HN-Kq"
      },
      "source": [
        "## Convert weights to ckpt to use in web UIs like AUTOMATIC1111."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "89Az5NUxOWdy"
      },
      "outputs": [],
      "source": [
        "#@markdown Run conversion.\n",
        "from natsort import natsorted\n",
        "from glob import glob\n",
        "import os\n",
        "WEIGHTS_DIR = natsorted(glob(OUTPUT_DIR + os.sep + \"*\"))[-1]\n",
        "print(f\"[*] WEIGHTS_DIR={WEIGHTS_DIR}\")\n",
        "\n",
        "ckpt_path = WEIGHTS_DIR + \"/model.ckpt\"\n",
        "\n",
        "half_arg = \"\"\n",
        "#@markdown  Whether to convert to fp16, takes half the space (2GB).\n",
        "fp16 = True #@param {type: \"boolean\"}\n",
        "if fp16:\n",
        "    half_arg = \"--half\"\n",
        "!python convert_diffusers_to_original_stable_diffusion.py --model_path $WEIGHTS_DIR  --checkpoint_path $ckpt_path $half_arg\n",
        "print(f\"[*] Converted ckpt saved at {ckpt_path}\")"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@title Save ckpt to google drive\n",
        "from google.colab import drive\n",
        "drive.mount('/content/drive')\n",
        "\n",
        "import os\n",
        "model_checkpoints = \"/content/drive/MyDrive/sd/stable-diffusion-webui/models/Stable-diffusion\"\n",
        "os.makedirs(model_checkpoints, exist_ok=True)\n",
        "!cp {ckpt_path} {model_checkpoints}"
      ],
      "metadata": {
        "cellView": "form",
        "id": "MYDjfXf8MB2R"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ToNG4fd_dTbF"
      },
      "source": [
        "## Inference"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "gW15FjffdTID"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "from torch import autocast\n",
        "from diffusers import StableDiffusionPipeline, DDIMScheduler\n",
        "from IPython.display import display\n",
        "\n",
        "model_path = WEIGHTS_DIR             # If you want to use previously trained model saved in gdrive, replace this with the full path of model in gdrive\n",
        "\n",
        "scheduler = DDIMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule=\"scaled_linear\", clip_sample=False, set_alpha_to_one=False)\n",
        "pipe = StableDiffusionPipeline.from_pretrained(model_path, scheduler=scheduler, safety_checker=None, torch_dtype=torch.float16).to(\"cuda\")\n",
        "\n",
        "g_cuda = None"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "oIzkltjpVO_f"
      },
      "outputs": [],
      "source": [
        "#@markdown Can set random seed here for reproducibility.\n",
        "g_cuda = torch.Generator(device='cuda')\n",
        "seed = 52362 #@param {type:\"number\"}\n",
        "g_cuda.manual_seed(seed)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "K6xoHWSsbcS3",
        "cellView": "form"
      },
      "outputs": [],
      "source": [
        "#@title Run for generating images.\n",
        "\n",
        "prompt = \"sks 1boy in a bucket\" #@param {type:\"string\"}\n",
        "negative_prompt = \"\" #@param {type:\"string\"}\n",
        "num_samples = 4 #@param {type:\"number\"}\n",
        "guidance_scale = 7.5 #@param {type:\"number\"}\n",
        "num_inference_steps = 50 #@param {type:\"number\"}\n",
        "height = 512 #@param {type:\"number\"}\n",
        "width = 512 #@param {type:\"number\"}\n",
        "\n",
        "with autocast(\"cuda\"), torch.inference_mode():\n",
        "    images = pipe(\n",
        "        prompt,\n",
        "        height=height,\n",
        "        width=width,\n",
        "        negative_prompt=negative_prompt,\n",
        "        num_images_per_prompt=num_samples,\n",
        "        num_inference_steps=num_inference_steps,\n",
        "        guidance_scale=guidance_scale,\n",
        "        generator=g_cuda\n",
        "    ).images\n",
        "\n",
        "for img in images:\n",
        "    display(img)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "WMCqQ5Tcdsm2"
      },
      "outputs": [],
      "source": [
        "#@markdown Run Gradio UI for generating images.\n",
        "import gradio as gr\n",
        "\n",
        "def inference(prompt, negative_prompt, num_samples, height=512, width=512, num_inference_steps=50, guidance_scale=7.5):\n",
        "    with torch.autocast(\"cuda\"), torch.inference_mode():\n",
        "        return pipe(\n",
        "                prompt, height=int(height), width=int(width),\n",
        "                negative_prompt=negative_prompt,\n",
        "                num_images_per_prompt=int(num_samples),\n",
        "                num_inference_steps=int(num_inference_steps), guidance_scale=guidance_scale,\n",
        "                generator=g_cuda\n",
        "            ).images\n",
        "\n",
        "with gr.Blocks() as demo:\n",
        "    with gr.Row():\n",
        "        with gr.Column():\n",
        "            prompt = gr.Textbox(label=\"Prompt\", value=\"photo of sks dog in a bucket\")\n",
        "            negative_prompt = gr.Textbox(label=\"Negative Prompt\", value=\"\")\n",
        "            run = gr.Button(value=\"Generate\")\n",
        "            with gr.Row():\n",
        "                num_samples = gr.Number(label=\"Number of Samples\", value=4)\n",
        "                guidance_scale = gr.Number(label=\"Guidance Scale\", value=7.5)\n",
        "            with gr.Row():\n",
        "                height = gr.Number(label=\"Height\", value=512)\n",
        "                width = gr.Number(label=\"Width\", value=512)\n",
        "            num_inference_steps = gr.Slider(label=\"Steps\", value=50)\n",
        "        with gr.Column():\n",
        "            gallery = gr.Gallery()\n",
        "\n",
        "    run.click(inference, inputs=[prompt, negative_prompt, num_samples, height, width, num_inference_steps, guidance_scale], outputs=gallery)\n",
        "\n",
        "demo.launch(debug=True)"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [
        "XfTlc8Mqb8iH"
      ],
      "provenance": [],
      "private_outputs": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3.8.12 ('pytorch')",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.8.12"
    },
    "vscode": {
      "interpreter": {
        "hash": "2d58e898dde0263bc564c6968b04150abacfd33eed9b19aaa8e45c040360e146"
      }
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
 }
	{
	"cells": [
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "view-in-github",
	"colab_type": "text"
	},
	"source": [
	"<a href=\"https://colab.research.google.com/gist/thx-pw/37ba770dfd2b15f119ec8032c3ae90a1/dreambooth_stable_diffusion_mod.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"cellView": "form",
	"id": "XU7NuMAA2drw"
	},
	"outputs": [],
	"source": [
	"#@markdown Check type of GPU and VRAM available.\n",
	"!nvidia-smi --query-gpu=name,memory.total,memory.free --format=csv,noheader"
	]
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "BzM7j0ZSc_9c"
	},
	"source": [
	"https://github.com/ShivamShrirao/diffusers/tree/main/examples/dreambooth"
	]
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "wnTMyW41cC1E"
	},
	"source": [
	"## Install Requirements"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"id": "aLWXPZqjsZVV"
	},
	"outputs": [],
	"source": [
	"!wget -q https://github.com/ShivamShrirao/diffusers/raw/main/examples/dreambooth/train_dreambooth.py\n",
	"!wget -q https://github.com/ShivamShrirao/diffusers/raw/main/scripts/convert_diffusers_to_original_stable_diffusion.py\n",
	"%pip install -qq git+https://github.com/ShivamShrirao/diffusers\n",
	"%pip install -q -U --pre triton\n",
	"%pip install -q accelerate transformers ftfy bitsandbytes==0.35.0 gradio natsort"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"cellView": "form",
	"id": "y4lqqWT_uxD2"
	},
	"outputs": [],
	"source": [
	"#@title Login to HuggingFace 🤗\n",
	"\n",
	"#@markdown You need to accept the model license before downloading or using the Stable Diffusion weights. Please, visit the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5), read the license and tick the checkbox if you agree. You have to be a registered user in 🤗 Hugging Face Hub, and you'll also need to use an access token for the code to work.\n",
	"# https://huggingface.co/settings/tokens\n",
	"!mkdir -p ~/.huggingface\n",
	"HUGGINGFACE_TOKEN = \"\" #@param {type:\"string\"}\n",
	"!echo -n \"{HUGGINGFACE_TOKEN}\" > ~/.huggingface/token"
	]
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "XfTlc8Mqb8iH"
	},
	"source": [
	"### Install xformers from precompiled wheel."
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"id": "n6dcjPnnaiCn"
	},
	"outputs": [],
	"source": [
	"%pip install --no-deps -q https://github.com/brian6091/xformers-wheels/releases/download/0.0.15.dev0%2B4c06c79/xformers-0.0.15.dev0+4c06c79.d20221205-cp38-cp38-linux_x86_64.whl\n",
	"# These were compiled on Tesla T4.\n",
	"\n",
	"# If precompiled wheels don't work, install it with the following command. It will take around 40 minutes to compile.\n",
	"# %pip install git+https://github.com/facebookresearch/xformers@4c06c79#egg=xformers"
	]
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "G0NV324ZcL9L"
	},
	"source": [
	"## Settings and run"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"id": "Rxg0y5MBudmd",
	"cellView": "form"
	},
	"outputs": [],
	"source": [
	"#@markdown Name/Path of the initial model.\n",
	"MODEL_NAME = \"hakurei/waifu-diffusion\" #@param [\"hakurei/waifu-diffusion\", \"runwayml/stable-diffusion-v1-5\", \"CompVis/stable-diffusion-v1-4\", \"naclbit/trinart_stable_diffusion_v2,diffusers-115k\", \"naclbit/trinart_stable_diffusion_v2,diffusers-95k\", \"naclbit/trinart_stable_diffusion_v2,diffusers-60k\"] {allow-input: true}\n",
	"\n",
	"INSTANCE_NAME = \"sks\" #@param {type:\"string\"}\n",
	"CLASS_NAME = \"1boy\" #@param {type:\"string\"}\n",
	"INSTANCE_PROMPT = f\"{INSTANCE_NAME} {CLASS_NAME}\"\n",
	"\n",
	"CLASS_DIR = f\"/content/data/{CLASS_NAME}\"\n",
	"\n",
	"INSTANCE_DIR = f\"/content/data/{INSTANCE_NAME}\"\n",
	"!mkdir -p $INSTANCE_DIR\n",
	"\n",
	"OUTPUT_DIR = f\"/content/stable_diffusion_weights/{INSTANCE_NAME}\"\n",
	"print(f\"[*] Weights will be saved at {OUTPUT_DIR}\")\n",
	"!mkdir -p $OUTPUT_DIR"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"cellView": "form",
	"id": "fe-GgtnUVO_e"
	},
	"outputs": [],
	"source": [
	"#@markdown Upload your images by running this cell.\n",
	"\n",
	"#@markdown OR\n",
	"\n",
	"#@markdown You can use the file manager on the left panel to upload (drag and drop) to INSTANCE_DIR (it uploads faster)\n",
	"\n",
	"import os\n",
	"from google.colab import files\n",
	"import shutil\n",
	"\n",
	"uploaded = files.upload()\n",
	"for filename in uploaded.keys():\n",
	" dst_path = os.path.join(INSTANCE_DIR, filename)\n",
	" shutil.move(filename, dst_path)"
	]
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "qn5ILIyDJIcX"
	},
	"source": [
	"# Start Training\n",
	"\n",
	"Use the table below to choose the best flags based on your memory and speed requirements. Tested on Tesla T4 GPU.\n",
	"\n",
	"\n",
	"\| `fp16` \| `train_batch_size` \| `gradient_accumulation_steps` \| `gradient_checkpointing` \| `use_8bit_adam` \| GB VRAM usage \| Speed (it/s) \|\n",
	"\| ---- \| ------------------ \| ----------------------------- \| ----------------------- \| --------------- \| ---------- \| ------------ \|\n",
	"\| fp16 \| 1 \| 1 \| TRUE \| TRUE \| 9.92 \| 0.93 \|\n",
	"\| no \| 1 \| 1 \| TRUE \| TRUE \| 10.08 \| 0.42 \|\n",
	"\| fp16 \| 2 \| 1 \| TRUE \| TRUE \| 10.4 \| 0.66 \|\n",
	"\| fp16 \| 1 \| 1 \| FALSE \| TRUE \| 11.17 \| 1.14 \|\n",
	"\| no \| 1 \| 1 \| FALSE \| TRUE \| 11.17 \| 0.49 \|\n",
	"\| fp16 \| 1 \| 2 \| TRUE \| TRUE \| 11.56 \| 1 \|\n",
	"\| fp16 \| 2 \| 1 \| FALSE \| TRUE \| 13.67 \| 0.82 \|\n",
	"\| fp16 \| 1 \| 2 \| FALSE \| TRUE \| 13.7 \| 0.83 \|\n",
	"\| fp16 \| 1 \| 1 \| TRUE \| FALSE \| 15.79 \| 0.77 \|\n"
	]
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "-ioxxvHoicPs"
	},
	"source": [
	"Add `--gradient_checkpointing` flag for around 9.92 GB VRAM usage.\n",
	"\n",
	"remove `--use_8bit_adam` flag for full precision. Requires 15.79 GB with `--gradient_checkpointing` else 17.8 GB."
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"id": "jjcSXTp-u-Eg"
	},
	"outputs": [],
	"source": [
	"!accelerate launch train_dreambooth.py \\\n",
	" --pretrained_model_name_or_path=$MODEL_NAME \\\n",
	" --pretrained_vae_name_or_path=\"\" \\\n",
	" --instance_data_dir=$INSTANCE_DIR \\\n",
	" --class_data_dir=$CLASS_DIR \\\n",
	" --output_dir=$OUTPUT_DIR \\\n",
	" --revision=\"fp16\" \\\n",
	" --with_prior_preservation --prior_loss_weight=1.0 \\\n",
	" --resolution=512 \\\n",
	" --train_batch_size=4 \\\n",
	" --train_text_encoder \\\n",
	" --pad_tokens \\\n",
	" --mixed_precision=\"fp16\" \\\n",
	" --gradient_checkpointing \\\n",
	" --use_8bit_adam \\\n",
	" --gradient_accumulation_steps=1 \\\n",
	" --learning_rate=1e-6 \\\n",
	" --lr_scheduler=\"constant\" \\\n",
	" --lr_warmup_steps=0 \\\n",
	" --num_class_images=60 \\\n",
	" --sample_batch_size=1 \\\n",
	" --max_train_steps=300 \\\n",
	" --instance_prompt=\"{INSTANCE_PROMPT}\" \\\n",
	" --class_prompt=\"{CLASS_NAME}\""
	]
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "5V8wgU0HN-Kq"
	},
	"source": [
	"## Convert weights to ckpt to use in web UIs like AUTOMATIC1111."
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"cellView": "form",
	"id": "89Az5NUxOWdy"
	},
	"outputs": [],
	"source": [
	"#@markdown Run conversion.\n",
	"from natsort import natsorted\n",
	"from glob import glob\n",
	"import os\n",
	"WEIGHTS_DIR = natsorted(glob(OUTPUT_DIR + os.sep + \"*\"))[-1]\n",
	"print(f\"[*] WEIGHTS_DIR={WEIGHTS_DIR}\")\n",
	"\n",
	"ckpt_path = WEIGHTS_DIR + \"/model.ckpt\"\n",
	"\n",
	"half_arg = \"\"\n",
	"#@markdown Whether to convert to fp16, takes half the space (2GB).\n",
	"fp16 = True #@param {type: \"boolean\"}\n",
	"if fp16:\n",
	" half_arg = \"--half\"\n",
	"!python convert_diffusers_to_original_stable_diffusion.py --model_path $WEIGHTS_DIR --checkpoint_path $ckpt_path $half_arg\n",
	"print(f\"[*] Converted ckpt saved at {ckpt_path}\")"
	]
	},
	{
	"cell_type": "code",
	"source": [
	"#@title Save ckpt to google drive\n",
	"from google.colab import drive\n",
	"drive.mount('/content/drive')\n",
	"\n",
	"import os\n",
	"model_checkpoints = \"/content/drive/MyDrive/sd/stable-diffusion-webui/models/Stable-diffusion\"\n",
	"os.makedirs(model_checkpoints, exist_ok=True)\n",
	"!cp {ckpt_path} {model_checkpoints}"
	],
	"metadata": {
	"cellView": "form",
	"id": "MYDjfXf8MB2R"
	},
	"execution_count": null,
	"outputs": []
	},
	{
	"cell_type": "markdown",
	"metadata": {
	"id": "ToNG4fd_dTbF"
	},
	"source": [
	"## Inference"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"id": "gW15FjffdTID"
	},
	"outputs": [],
	"source": [
	"import torch\n",
	"from torch import autocast\n",
	"from diffusers import StableDiffusionPipeline, DDIMScheduler\n",
	"from IPython.display import display\n",
	"\n",
	"model_path = WEIGHTS_DIR # If you want to use previously trained model saved in gdrive, replace this with the full path of model in gdrive\n",
	"\n",
	"scheduler = DDIMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule=\"scaled_linear\", clip_sample=False, set_alpha_to_one=False)\n",
	"pipe = StableDiffusionPipeline.from_pretrained(model_path, scheduler=scheduler, safety_checker=None, torch_dtype=torch.float16).to(\"cuda\")\n",
	"\n",
	"g_cuda = None"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"cellView": "form",
	"id": "oIzkltjpVO_f"
	},
	"outputs": [],
	"source": [
	"#@markdown Can set random seed here for reproducibility.\n",
	"g_cuda = torch.Generator(device='cuda')\n",
	"seed = 52362 #@param {type:\"number\"}\n",
	"g_cuda.manual_seed(seed)"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"id": "K6xoHWSsbcS3",
	"cellView": "form"
	},
	"outputs": [],
	"source": [
	"#@title Run for generating images.\n",
	"\n",
	"prompt = \"sks 1boy in a bucket\" #@param {type:\"string\"}\n",
	"negative_prompt = \"\" #@param {type:\"string\"}\n",
	"num_samples = 4 #@param {type:\"number\"}\n",
	"guidance_scale = 7.5 #@param {type:\"number\"}\n",
	"num_inference_steps = 50 #@param {type:\"number\"}\n",
	"height = 512 #@param {type:\"number\"}\n",
	"width = 512 #@param {type:\"number\"}\n",
	"\n",
	"with autocast(\"cuda\"), torch.inference_mode():\n",
	" images = pipe(\n",
	" prompt,\n",
	" height=height,\n",
	" width=width,\n",
	" negative_prompt=negative_prompt,\n",
	" num_images_per_prompt=num_samples,\n",
	" num_inference_steps=num_inference_steps,\n",
	" guidance_scale=guidance_scale,\n",
	" generator=g_cuda\n",
	" ).images\n",
	"\n",
	"for img in images:\n",
	" display(img)"
	]
	},
	{
	"cell_type": "code",
	"execution_count": null,
	"metadata": {
	"cellView": "form",
	"id": "WMCqQ5Tcdsm2"
	},
	"outputs": [],
	"source": [
	"#@markdown Run Gradio UI for generating images.\n",
	"import gradio as gr\n",
	"\n",
	"def inference(prompt, negative_prompt, num_samples, height=512, width=512, num_inference_steps=50, guidance_scale=7.5):\n",
	" with torch.autocast(\"cuda\"), torch.inference_mode():\n",
	" return pipe(\n",
	" prompt, height=int(height), width=int(width),\n",
	" negative_prompt=negative_prompt,\n",
	" num_images_per_prompt=int(num_samples),\n",
	" num_inference_steps=int(num_inference_steps), guidance_scale=guidance_scale,\n",
	" generator=g_cuda\n",
	" ).images\n",
	"\n",
	"with gr.Blocks() as demo:\n",
	" with gr.Row():\n",
	" with gr.Column():\n",
	" prompt = gr.Textbox(label=\"Prompt\", value=\"photo of sks dog in a bucket\")\n",
	" negative_prompt = gr.Textbox(label=\"Negative Prompt\", value=\"\")\n",
	" run = gr.Button(value=\"Generate\")\n",
	" with gr.Row():\n",
	" num_samples = gr.Number(label=\"Number of Samples\", value=4)\n",
	" guidance_scale = gr.Number(label=\"Guidance Scale\", value=7.5)\n",
	" with gr.Row():\n",
	" height = gr.Number(label=\"Height\", value=512)\n",
	" width = gr.Number(label=\"Width\", value=512)\n",
	" num_inference_steps = gr.Slider(label=\"Steps\", value=50)\n",
	" with gr.Column():\n",
	" gallery = gr.Gallery()\n",
	"\n",
	" run.click(inference, inputs=[prompt, negative_prompt, num_samples, height, width, num_inference_steps, guidance_scale], outputs=gallery)\n",
	"\n",
	"demo.launch(debug=True)"
	]
	}
	],
	"metadata": {
	"accelerator": "GPU",
	"colab": {
	"collapsed_sections": [
	"XfTlc8Mqb8iH"
	],
	"provenance": [],
	"private_outputs": true,
	"include_colab_link": true
	},
	"kernelspec": {
	"display_name": "Python 3.8.12 ('pytorch')",
	"language": "python",
	"name": "python3"
	},
	"language_info": {
	"name": "python",
	"version": "3.8.12"
	},
	"vscode": {
	"interpreter": {
	"hash": "2d58e898dde0263bc564c6968b04150abacfd33eed9b19aaa8e45c040360e146"
	}
	}
	},
	"nbformat": 4,
	"nbformat_minor": 0
	}