Upload Initial.ipynb

Initial.ipynb

@@ -0,0 +1,338 @@
+{
+  "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"
+          ]
+        }
+      ]
+    }
+  ]
+}