File size: 5,159 Bytes
db14b86
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
{
 "cells": [
  {
   "cell_type": "markdown",
   "source": [
    "## Setup\n",
    "\n",
    "Execute the cell below to setup the notebook to run Custom diffusion"
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "source": [
    "!bash setup.sh "
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Training\n",
    "\n",
    "## Diffusers method"
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "source": [
    "## launch training script (2 GPUs recommended, increase --max_train_steps to 500 if 1 GPU)\n",
    "\n",
    "!accelerate launch src/diffuser_training.py \\\n",
    "          --pretrained_model_name_or_path=compvis/stable-diffusion-v1-4   \\\n",
    "          --instance_data_dir=./data/cat  \\\n",
    "          --class_data_dir=./real_reg/samples_cat/ \\\n",
    "          --output_dir=./logs/cat  \\\n",
    "          --with_prior_preservation --real_prior --prior_loss_weight=1.0 \\\n",
    "          --instance_prompt=\"photo of a <new1> cat\"  \\\n",
    "          --class_prompt=\"cat\" \\\n",
    "          --resolution=512  \\\n",
    "          --train_batch_size=2  \\\n",
    "          --learning_rate=1e-5  \\\n",
    "          --lr_warmup_steps=0 \\\n",
    "          --max_train_steps=250 \\\n",
    "          --num_class_images=200 \\\n",
    "          --scale_lr \\\n",
    "          --modifier_token \"<new1>\"\n",
    "\n",
    "## sample \n",
    "python src/sample_diffuser.py --delta_ckpt logs/cat/delta.bin --ckpt \"CompVis/stable-diffusion-v1-4\" --prompt \"<new1> cat playing with a ball\"\n"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "source": [
    "## launch training script (2 GPUs recommended, increase --max_train_steps to 1000 if 1 GPU)\n",
    "## provide some json file with the info about each concept\n",
    "!CUDA_VISIBLE_DEVICES=2,3 accelerate launch src/diffuser_training.py \\\n",
    "          --pretrained_model_name_or_path=compvis/stable-diffusion-v1-4  \\\n",
    "          --output_dir=./logs/cat_wooden_pot  \\\n",
    "          --concepts_list=./assets/concept_list.json \\\n",
    "          --with_prior_preservation --real_prior --prior_loss_weight=1.0 \\\n",
    "          --resolution=512  \\\n",
    "          --train_batch_size=2  \\\n",
    "          --learning_rate=1e-5  \\\n",
    "          --lr_warmup_steps=0 \\\n",
    "          --max_train_steps=500 \\\n",
    "          --num_class_images=200 \\\n",
    "          --scale_lr --hflip  \\\n",
    "          --modifier_token \"<new1>+<new2>\" \n",
    "\n",
    "## sample \n",
    "python src/sample_diffuser.py --delta_ckpt logs/cat_wooden_pot/delta.bin --ckpt \"CompVis/stable-diffusion-v1-4\" --prompt \"<new1> cat sitting inside a <new2> wooden pot and looking up\""
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Gradio application"
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "source": [
    "%cd ~/../custom-diffusion\n",
    "!python app.py --share"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Sample from the newly trained model"
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "source": [
    "import torch\n",
    "from torch import autocast\n",
    "from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler\n",
    "\n",
    "%cd ~/../notebooks\n",
    "\n",
    "## Set the args\n",
    "device = 'cuda'\n",
    "model_ = 'custom-diffusion/results/checkpoint-500'\n",
    "size_ = 512\n",
    "precision = 512\n",
    "sample_num = 5\n",
    "\n",
    "print(f'Generating samples from Stable Diffusion {model_} checkpoint ({precision})')\n",
    "\n",
    "\n",
    "## Instantiate the model pipe with StableDiffusionPipeline\n",
    "pipe = StableDiffusionPipeline.from_pretrained(model_, torch_dtype=torch.float16, revision=\"fp16\")\n",
    "pipe = pipe.to(device)\n",
    "\n",
    "for j in range(sample_num):\n",
    "    a = pipe(prompt = 'a photo of a <new1> cat',\n",
    "             negative_prompt = None,\n",
    "             guidance_scale=7.5,\n",
    "             height = 512,\n",
    "             width = 512,\n",
    "             num_images_per_prompt = 1,\n",
    "             num_inference_steps=50)['images']    \n",
    "    for i in a:\n",
    "        display(i)\n",
    "    # hash the next line out to save results\n",
    "    #a.save(f'outputs/gen-image-{j}.png')\n"
   ],
   "outputs": [],
   "metadata": {}
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Compress the model"
   ],
   "metadata": {}
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "source": [
    "!python src/compress.py --delta_ckpt <finetuned-delta-path> --ckpt <pretrained-model-path>"
   ],
   "outputs": [],
   "metadata": {}
  }
 ],
 "metadata": {
  "orig_nbformat": 4,
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}