File size: 13,494 Bytes
b8bb788 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 |
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"provenance": []
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"language_info": {
"name": "python"
},
"accelerator": "TPU"
},
"cells": [
{
"cell_type": "code",
"source": [
"!pip install git+https://github.com/huggingface/diffusers.git\n",
"!pip install -U -r requirements.txt\n",
"!pip install huggingface\n",
"!pip install diffusers[training]\n",
"!pip install diffusers\n",
"!pip install torch\n",
"!sudo apt -qq install git-lfs\n",
"!git config --global credential.helper store\n",
"!pip install tqdm"
],
"metadata": {
"id": "aE5NZ-XcU7bC"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"from dataclasses import dataclass\n",
"\n",
"\n",
"@dataclass\n",
"class TrainingConfig:\n",
" image_size = 128 # the generated image resolution\n",
" train_batch_size = 16\n",
" eval_batch_size = 16 # how many images to sample during evaluation\n",
" num_epochs = 50\n",
" gradient_accumulation_steps = 1\n",
" learning_rate = 1e-4\n",
" lr_warmup_steps = 500\n",
" save_image_epochs = 10\n",
" save_model_epochs = 10\n",
" mixed_precision = \"fp16\" # `no` for float32, `fp16` for automatic mixed precision\n",
" output_dir = \"ddpm-butterflies-128\" # the model name locally and on the HF Hub\n",
"\n",
" push_to_hub = True # whether to upload the saved model to the HF Hub\n",
" hub_private_repo = False\n",
" overwrite_output_dir = True # overwrite the old model when re-running the notebook\n",
" seed = 0\n",
"\n",
"\n",
"config = TrainingConfig()"
],
"metadata": {
"id": "faBx8T9NV1Xv"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "d5jOnnaPSKZx"
},
"outputs": [],
"source": [
"from datasets import load_dataset\n",
"\n",
"config.dataset_name = \"Drozdik/tattoo_v0\"\n",
"dataset = load_dataset(config.dataset_name, split=\"train\")"
]
},
{
"cell_type": "code",
"source": [
"def transform(examples):\n",
" images = [preprocess(image.convert(\"RGB\")) for image in examples[\"image\"]]\n",
" return {\"images\": images}\n",
"\n"
],
"metadata": {
"id": "CvUPjQmqXsG1"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"from diffusers import DDPMPipeline\n",
"import math\n",
"import os\n",
"\n",
"def make_grid(images, rows, cols):\n",
" w, h = images[0].size\n",
" grid = Image.new(\"RGB\", size=(cols * w, rows * h))\n",
" for i, image in enumerate(images):\n",
" grid.paste(image, box=(i % cols * w, i // cols * h))\n",
" return grid\n",
"\n",
"\n",
"def evaluate(config, epoch, pipeline):\n",
" images = pipeline(\n",
" batch_size=config.eval_batch_size,\n",
" generator=torch.manual_seed(config.seed),\n",
" ).images\n",
"\n",
" image_grid = make_grid(images, rows=4, cols=4)\n",
"\n",
" test_dir = os.path.join(config.output_dir, \"samples\")\n",
" os.makedirs(test_dir, exist_ok=True)\n",
" image_grid.save(f\"{test_dir}/{epoch:04d}.png\")\n",
"\n"
],
"metadata": {
"id": "p6tO2qgGx-m3"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"from accelerate import Accelerator\n",
"from tqdm.auto import tqdm\n",
"from pathlib import Path\n",
"import os\n",
"\n",
"def train_loop(config, model, noise_scheduler, optimizer, train_dataloader, lr_scheduler):\n",
" accelerator = Accelerator(\n",
" mixed_precision=config.mixed_precision,\n",
" gradient_accumulation_steps=config.gradient_accumulation_steps,\n",
" log_with=\"tensorboard\",\n",
" project_dir=os.path.join(config.output_dir, \"logs\"),\n",
" )\n",
"\n",
" if accelerator.is_main_process:\n",
" os.makedirs(config.output_dir,exist_ok=True)\n",
" accelerator.init_trackers(\"train_example\")\n",
"\n",
" model, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(model, optimizer, train_dataloader, lr_scheduler)\n",
"\n",
" global_step = 0\n",
"\n",
" for epoch in range(config.num_epochs):\n",
" progress_bar = tqdm(total=len(train_dataloader), disable=not accelerator.is_local_main_process)\n",
" progress_bar.set_description(f\"Epoch {epoch}\")\n",
"\n",
" for step, batch in enumerate(train_dataloader):\n",
" clean_images = batch[\"images\"]\n",
"\n",
" noise = torch.randn(clean_images.shape).to(clean_images.device)\n",
" bs = clean_images.shape[0]\n",
"\n",
" timesteps = torch.randint(\n",
" 0, noise_scheduler.config.num_train_timesteps, (bs,), device=clean_images.device\n",
" ).long()\n",
"\n",
" noisy_images = noise_scheduler.add_noise(clean_images, noise, timesteps)\n",
"\n",
" with accelerator.accumulate(model):\n",
" noise_pred = model(noisy_images, timesteps, return_dict=False)[0]\n",
" loss = F.mse_loss(noise_pred,noise)\n",
" accelerator.backward(loss)\n",
"\n",
" accelerator.clip_grad_norm_(model.parameters(),1.0)\n",
" optimizer.step()\n",
" lr_scheduler.step()\n",
" optimizer.zero_grad()\n",
"\n",
" progress_bar.update(1)\n",
" logs = {\"loss\": loss.detach().item(), \"lr\": lr_scheduler.get_last_lr()[0], \"step\": global_step}\n",
" progress_bar.set_postfix(**logs)\n",
" accelerator.log(logs, step=global_step)\n",
" global_step += 1\n",
"\n",
" if accelerator.is_main_process:\n",
" pipeline = DDPMPipeline(unet=accelerator.unwrap_model(model), scheduler=noise_scheduler)\n",
"\n",
" if (epoch + 1) % config.save_image_epochs == 0 or epoch == config.num_epochs - 1:\n",
" evaluate(config, epoch, pipeline)\n",
" if (epoch + 1) % config.save_model_epochs == 0 or epoch == config.num_epochs - 1:\n",
" pipeline.save_pretrained(config.output_dir)\n",
"\n",
"\n",
"\n",
" upload_folder(\n",
" repo_id=repo_id,\n",
" folder_path=args.output_dir,\n",
" commit_message=\"End of training\",\n",
" ignore_patterns=[\"step_*\", \"epoch_*\"],\n",
" )\n",
"\n"
],
"metadata": {
"id": "Ae7g7TaCsnh7"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"from accelerate import notebook_launcher\n",
"import torch.nn.functional as F\n",
"from diffusers.optimization import get_cosine_schedule_with_warmup\n",
"import torch\n",
"from PIL import Image\n",
"from diffusers import DDPMScheduler\n",
"from diffusers import UNet2DModel\n",
"import torch\n",
"from torchvision import transforms\n",
"\n",
"\n",
"\n",
"\n",
"preprocess = transforms.Compose(\n",
" [\n",
" transforms.Resize((config.image_size, config.image_size)),\n",
" transforms.RandomHorizontalFlip(),\n",
" transforms.ToTensor(),\n",
" transforms.Normalize([.5],[.5]),\n",
" ]\n",
")\n",
"\n",
"dataset.set_transform(transform)\n",
"\n",
"train_dataloader = torch.utils.data.DataLoader(dataset, batch_size=config.train_batch_size, shuffle=True)\n",
"\n",
"model = UNet2DModel(sample_size=config.image_size,in_channels=3, out_channels=3, layers_per_block=2, block_out_channels=(128,128,256,256,512,512), down_block_types=(\"DownBlock2D\",\"DownBlock2D\",\"DownBlock2D\",\"DownBlock2D\",\"AttnDownBlock2D\",\"DownBlock2D\"), up_block_types=(\"UpBlock2D\",\"AttnUpBlock2D\",\"UpBlock2D\",\"UpBlock2D\",\"UpBlock2D\",\"UpBlock2D\"), )\n",
"\n",
"sample_image = dataset[0][\"images\"].unsqueeze(0)\n",
"print(\"Input shape:\", sample_image.shape)\n",
"\n",
"print(\"Output shape:\", model(sample_image, timestep=0).sample.shape)\n",
"\n",
"noise_scheduler = DDPMScheduler(num_train_timesteps=1000)\n",
"noise = torch.randn(sample_image.shape)\n",
"time_steps = torch.LongTensor([50])\n",
"noisy_image = noise_scheduler.add_noise(sample_image, noise, time_steps)\n",
"Image.fromarray(((noisy_image.permute(0, 2, 3, 1) + 1.0) * 127.5).type(torch.uint8).numpy()[0])\n",
"\n",
"\n",
"\n",
"\n",
"optimizer = torch.optim.AdamW(model.parameters(), lr=config.learning_rate)\n",
"lr_scheduler = get_cosine_schedule_with_warmup(\n",
" optimizer=optimizer,\n",
" num_warmup_steps=config.lr_warmup_steps,\n",
" num_training_steps=(len(train_dataloader)*config.num_epochs),\n",
")\n",
"\n",
"\n",
"\n",
"args = (config, model, noise_scheduler, optimizer, train_dataloader, lr_scheduler)\n",
"\n",
"notebook_launcher(train_loop, args, num_processes=1)"
],
"metadata": {
"id": "FnPpL7H2yT8O"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"model = UNet2DModel.from_pretrained(config.output_dir, subfolder=\"unet\")\n",
"optimizer = torch.optim.AdamW(model.parameters(), lr=config.learning_rate)\n",
"lr_scheduler = get_cosine_schedule_with_warmup(\n",
" optimizer=optimizer,\n",
" num_warmup_steps=config.lr_warmup_steps,\n",
" num_training_steps=(len(train_dataloader)*config.num_epochs),\n",
")\n",
"args = (config, model, noise_scheduler, optimizer, train_dataloader, lr_scheduler)\n",
"notebook_launcher(train_loop, args, num_processes=1)"
],
"metadata": {
"id": "K22cx-8snBIV"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"!nvidia-smi"
],
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "Rqv9HTR22qXe",
"outputId": "9480fd9d-5545-4ef8-f91c-f1dc8a02573a"
},
"execution_count": null,
"outputs": [
{
"output_type": "stream",
"name": "stdout",
"text": [
"Sun Aug 6 08:13:38 2023 \n",
"+-----------------------------------------------------------------------------+\n",
"| NVIDIA-SMI 525.105.17 Driver Version: 525.105.17 CUDA Version: 12.0 |\n",
"|-------------------------------+----------------------+----------------------+\n",
"| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |\n",
"| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |\n",
"| | | MIG M. |\n",
"|===============================+======================+======================|\n",
"| 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 |\n",
"| N/A 77C P0 34W / 70W | 10807MiB / 15360MiB | 0% Default |\n",
"| | | N/A |\n",
"+-------------------------------+----------------------+----------------------+\n",
" \n",
"+-----------------------------------------------------------------------------+\n",
"| Processes: |\n",
"| GPU GI CI PID Type Process name GPU Memory |\n",
"| ID ID Usage |\n",
"|=============================================================================|\n",
"+-----------------------------------------------------------------------------+\n"
]
}
]
}
]
} |