{"id":217,"date":"2022-08-30T10:40:20","date_gmt":"2022-08-30T01:40:20","guid":{"rendered":"https:\/\/sheltie-garage.xyz\/tech\/?p=217"},"modified":"2022-08-30T11:00:17","modified_gmt":"2022-08-30T02:00:17","slug":"intel-mac%e3%81%a7stable-diffusion%e3%82%92%e5%8b%95%e3%81%8b%e3%81%99","status":"publish","type":"post","link":"https:\/\/sheltie-garage.xyz\/tech\/2022\/08\/intel-mac%e3%81%a7stable-diffusion%e3%82%92%e5%8b%95%e3%81%8b%e3%81%99\/","title":{"rendered":"Intel Mac\u3067Stable diffusion\u3092\u52d5\u304b\u3059"},"content":{"rendered":"\n

\u8a71\u984c\u306e\u304a\u7d75\u304b\u304dAI\u300cStable Diffusion\u300d\u3092Intel Mac\u3067\u52d5\u304b\u3057\u307e\u3059\u3002
\u3059\u3067\u306b\u5b9f\u8df5\u3055\u308c\u3066\u3044\u308b\u65b9\u304c\u3044\u3089\u3057\u305f\u306e\u3067\u3001\u4ee5\u4e0b\u306e\u8a18\u4e8b\u3092\u53c2\u8003\u306b\u3055\u305b\u3066\u3044\u305f\u3060\u304d\u307e\u3057\u305f\u3002
https:\/\/yaritakunai.hatenablog.com\/entry\/stable-diffusion-mps<\/a><\/p>\n\n\n\n

Intel Mac\u3067\u52d5\u304b\u3059\u3068\u304d\u306e\u554f\u984c\u70b9<\/h2>\n\n\n\n

Stable Diffusion\u3067\u3059\u304c\u3001\u516c\u5f0f\u306e\u5c0e\u5165\u624b\u9806\u3067\u306fCuda\u3092\u5229\u7528\u3059\u308b\u8a2d\u5b9a\u3068\u306a\u3063\u3066\u304a\u308a\u3001Ver 11.3\u304c\u8981\u6c42\u3055\u308c\u307e\u3059\u3002
\u305f\u3060\u3001Mac\u7528\u306eCuda\u306fVer9\u3067\u958b\u767a\u304c\u7d42\u4e86\u3057\u3066\u304a\u308a\u3001\u74b0\u5883\u69cb\u7bc9\u304c\u884c\u306a\u3048\u307e\u305b\u3093\u3002
Mac\u7528Cuda\u306f\u958b\u767a\u7d42\u4e86\u3068\u516c\u5f0f\u3067\u767a\u8868\u3082\u3042\u308a\u307e\u3057\u305f\u3002
https:\/\/applech2.com\/archives\/20200606-nvidia-cuda-end-of-support-macos.html<\/a><\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Cuda\u3092\u5229\u7528\u3057\u306a\u3044\u3088\u3046\u306b\u8a2d\u5b9a\u5909\u66f4\u3057\u3001Stable Diffusion\u306e\u74b0\u5883\u3092\u69cb\u7bc9\u3059\u308b<\/h2>\n\n\n\n

\u74b0\u5883\u306b\u3064\u3044\u3066\u306f\u3001\u4ee5\u4e0b\u306e\u901a\u308a<\/p>\n\n\n\n

\u30fbMacBook Pro 16inch 2019
\u30fbmacOS Monterey 12.5.1
\u30fbconda version : 4.10.3<\/p>\n\n\n\n

Gti\u304b\u3089\u30ea\u30dd\u30b8\u30c8\u30ea\u53d6\u5f97<\/h3>\n\n\n\n

\u4ee5\u4e0b\u306e\u30b3\u30de\u30f3\u30c9\u3067\u30ea\u30dd\u30b8\u30c8\u30ea\u3092\u53d6\u5f97\u3057\u307e\u3059<\/p>\n\n\n\n

git clone https:\/\/github.com\/CompVis\/stable-diffusion.git\ncd stable-diffusion<\/code><\/pre><\/div>\n\n\n\n
Cuda\u3092\u5229\u7528\u3057\u306a\u3044\u3088\u3046\u306b\u5909\u66f4<\/h3>\n\n\n\n
\u8a2d\u5b9a\u30d5\u30a1\u30a4\u30eb\u3092\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u5909\u66f4\u3057\u307e\u3059<\/p>\n\n\n\n
vi environment.yaml\n\n\u2193environment.yaml\u306e\u4e2d\u8eab\nname: ldm\nchannels:\n#  - pytorch\n  - pytorch-nightly\n  - defaults\ndependencies:\n  - python=3.8.5\n  - pip=20.3\n#  - cudatoolkit=11.3\n#  - pytorch=1.11.0\n#  - torchvision=0.12.0\n  - pytorch\n  - torchvision\n  - numpy=1.19.2\n  - pip:\n\n\u4ee5\u4e0b\u5909\u66f4\u306a\u3057<\/code><\/pre><\/div>\n\n\n\n
\u4eee\u60f3\u74b0\u5883\u4f5c\u6210<\/p>\n\n\n\n
\u4ee5\u4e0b\u306e\u30b3\u30de\u30f3\u30c9\u3067\u4eee\u60f3\u74b0\u5883\u3092\u4f5c\u6210\u3057\u3001\u6709\u52b9\u5316\u3057\u307e\u3059<\/p>\n\n\n\n
conda env create -f environment.yaml\nconda activate ldm\nmkdir -p models\/ldm\/stable-diffusion-v1<\/code><\/pre><\/div>\n\n\n\n
\u5b66\u7fd2\u30e2\u30c7\u30eb\u3092\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3059\u308b<\/h3>\n\n\n\n
Hugging Face<\/a>\u3068\u3044\u3046\u30b5\u30a4\u30c8\u304b\u3089\u5b66\u7fd2\u30c7\u30fc\u30bf\u3092\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u307e\u3059(\u8981\u4f1a\u54e1\u767b\u9332)
\u4f1a\u54e1\u767b\u9332\u5f8c\u3001\u4ee5\u4e0b\u306eURL\u304b\u3089\u30e2\u30c7\u30eb\u30c7\u30fc\u30bf\u3092\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u3001\u5148\u307b\u3069\u4f5c\u6210\u3057\u305f\u300cstable-diffusion-v1\u300d\u30d5\u30a9\u30eb\u30c0\u306b\u4fdd\u5b58\u3057\u307e\u3059
https:\/\/huggingface.co\/CompVis\/stable-diffusion-v-1-4-original<\/a><\/p>\n\n\n\n
\u4e0a\u8a18URL\u306b\u30a2\u30af\u30bb\u30b9\u5f8c\u3001\u30e9\u30a4\u30bb\u30f3\u30b9\u306b\u540c\u610f\u3057\u3001\u30e2\u30c7\u30eb\u3092\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3057\u307e\u3059<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u5f8c\u3001\u30d5\u30a1\u30a4\u30eb\u540d\u3092\u300cmodel.ckpt\u300d\u306b\u5909\u66f4\u3057\u307e\u3059<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3092\u5b9f\u884c\u3057\u3066\u307f\u308b<\/h3>\n\n\n\n
\u4ee5\u4e0b\u306epython\u30d5\u30a1\u30a4\u30eb\u3092\u4f5c\u6210\u3057\u3001\u5b9f\u884c\u3057\u3066\u307f\u307e\u3059\u3002\u30d5\u30a1\u30a4\u30eb\u306f\u300cstable-diffusion\u300d\u30d5\u30a9\u30eb\u30c0\u306e\u76f4\u4e0b\u306b\u4f5c\u6210\u3057\u307e\u3057\u305f
(\u30d7\u30ed\u30b0\u30e9\u30e0\u306f\u53c2\u8003\u8a18\u4e8b\u304b\u3089\u306e\u5f15\u7528\u306b\u306a\u308a\u307e\u3059)<\/p>\n\n\n\n
pytorch_m1_macbook.py<\/strong><\/p>\n\n\n\n
# -*- coding: utf-8 -*-\nimport torch\nfrom torch.nn import CrossEntropyLoss\nfrom torch.optim import SGD\nfrom torch.utils.data import DataLoader\nfrom torchvision.datasets import CIFAR10\nfrom torchvision import transforms as tt\nfrom torchvision.models import resnet18\n\nimport os\nfrom argparse import ArgumentParser\nimport time\n\ndef main(device):\n    # ResNet\u306e\u30cf\u30a4\u30d1\u30fc\u30d1\u30e9\u30e1\u30fc\u30bf\n    n_epoch = 5            # \u30a8\u30dd\u30c3\u30af\u6570\n    batch_size = 512       # \u30df\u30cb\u30d0\u30c3\u30c1\u30b5\u30a4\u30ba\n    momentum = 0.9         # SGD\u306emomentum\n    lr = 0.01              # \u5b66\u7fd2\u7387\n    weight_decay = 0.00005 # weight decay\n\n    # \u8a13\u7df4\u30c7\u30fc\u30bf\u3068\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u3092\u7528\u610f\n    mean = (0.491, 0.482, 0.446)\n    std = (0.247, 0.243, 0.261)\n    train_transform = tt.Compose([\n        tt.RandomHorizontalFlip(p=0.5),\n        tt.RandomCrop(size=32, padding=4, padding_mode='reflect'),\n        tt.ToTensor(),\n        tt.Normalize(mean=mean, std=std)\n    ])\n    test_transform = tt.Compose([tt.ToTensor(), tt.Normalize(mean, std)])\n    root = os.path.dirname(os.path.abspath(__file__))\n    train_set = CIFAR10(root=root, train=True,\n                        download=True, transform=train_transform)\n    train_loader = DataLoader(train_set, batch_size=batch_size,\n                              shuffle=True, num_workers=8)\n\n    # ResNet\u306e\u6e96\u5099\n    resnet = resnet18()\n    resnet.fc = torch.nn.Linear(resnet.fc.in_features, 10)\n\n    # \u8a13\u7df4\n    criterion = CrossEntropyLoss()\n    optimizer = SGD(resnet.parameters(), lr=lr,\n                    momentum=momentum, weight_decay=weight_decay)\n    train_start_time = time.time()\n    resnet.to(device)\n    resnet.train()\n    for epoch in range(1, n_epoch+1):\n        train_loss = 0.0\n        for inputs, labels in train_loader:\n            inputs = inputs.to(device)\n            optimizer.zero_grad()\n            outputs = resnet(inputs)\n            labels = labels.to(device)\n            loss = criterion(outputs, labels)\n            loss.backward()\n            train_loss += loss.item()\n            del loss  # \u30e1\u30e2\u30ea\u7bc0\u7d04\u306e\u305f\u3081\n            optimizer.step()\n        print('Epoch {} \/ {}: time = {}[s], loss = {:.2f}'.format(\n            epoch, n_epoch, time.time() - train_start_time, train_loss))\n    print('Train time on {}: {:.2f}[s] (Train loss = {:.2f})'.format(\n        device, time.time() - train_start_time, train_loss))\n\n    # \u8a55\u4fa1\n    test_set = CIFAR10(root=root, train=False, download=True,\n                       transform=test_transform)\n    test_loader = DataLoader(test_set, batch_size=batch_size,\n                             shuffle=False, num_workers=8)\n    test_loss = 0.0\n    test_start_time = time.time()\n    resnet.eval()\n    for inputs, labels in test_loader:\n        inputs = inputs.to(device)\n        outputs = resnet(inputs)\n        labels = labels.to(device)\n        loss = criterion(outputs, labels)\n        test_loss += loss.item()\n    print('Test time on {}: {:.2f}[s](Test loss = {:.2f})'.format(\n        device, time.time() - test_start_time, test_loss))\n\n\nif __name__ == '__main__':\n    parser = ArgumentParser()\n    parser.add_argument('--device', type=str, default='mps',\n                        choices=['cpu', 'mps'])\n    args = parser.parse_args()\n    device = torch.device(args.device)\n    main(device)<\/code><\/pre><\/div>\n\n\n\n
CPU\u3001MPS\u3067\u305d\u308c\u5b9f\u884c\u3057\u305f\u7d50\u679c\u306f\u4ee5\u4e0b\u306e\u3068\u304a\u308a\u3067\u3059<\/p>\n\n\n\n
python pytorch_m1_macbook.py --device cpu\nDownloading https:\/\/www.cs.toronto.edu\/~kriz\/cifar-10-python.tar.gz to \/Users\/hamada\/Project\/MachineLarning\/stable-diffusion\/cifar-10-python.tar.gz\n100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 170498071\/170498071 [00:22<00:00, 7461526.71it\/s]\nExtracting \/Users\/hamada\/Project\/MachineLarning\/stable-diffusion\/cifar-10-python.tar.gz to \/Users\/hamada\/Project\/MachineLarning\/stable-diffusion\nEpoch 1 \/ 5: time = 323.73796010017395[s], loss = 169.02\nEpoch 2 \/ 5: time = 628.1439380645752[s], loss = 135.66\nEpoch 3 \/ 5: time = 929.2913389205933[s], loss = 122.78\nEpoch 4 \/ 5: time = 1244.1902849674225[s], loss = 112.50\nEpoch 5 \/ 5: time = 1558.1003341674805[s], loss = 104.66\nTrain time on cpu: 1558.10[s] (Train loss = 104.66)\nFiles already downloaded and verified\nTest time on cpu: 64.05[s](Test loss = 20.15)<\/code><\/pre><\/div>\n\n\n\n
python pytorch_m1_macbook.py --device mps\nFiles already downloaded and verified\nEpoch 1 \/ 5: time = 243.6470718383789[s], loss = 170.26\nEpoch 2 \/ 5: time = 408.21199798583984[s], loss = 137.86\nEpoch 3 \/ 5: time = 625.620924949646[s], loss = 123.72\nEpoch 4 \/ 5: time = 857.4977359771729[s], loss = 113.30\nEpoch 5 \/ 5: time = 1090.7507467269897[s], loss = 105.25\nTrain time on mps: 1090.75[s] (Train loss = 105.25)\nFiles already downloaded and verified\nTest time on mps: 64.08[s](Test loss = 20.00)<\/code><\/pre><\/div>\n\n\n\n
mps\u3092\u5229\u7528\u3057\u305f\u307b\u3046\u304c30%\u307b\u3069\u9ad8\u901f\u306b\u51e6\u7406\u3067\u304d\u308b\u3088\u3046\u3067\u3059
\u4e0a\u8a18\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u5b9f\u884c\u7d50\u679c\u3060\u3051\u898b\u308b\u3068mps\u304c\u5229\u7528\u3067\u304d\u305d\u3046\u3067\u3059\u304c\u3001\u5b9f\u969b\u306b\u30a4\u30e9\u30b9\u30c8\u751f\u6210\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u8d77\u52d5\u3059\u308b\u3068\u30a8\u30e9\u30fc\u304c\u51fa\u3066\u51e6\u7406\u304c\u4e2d\u65ad\u3055\u308c\u3066\u3057\u307e\u3044\u307e\u3057\u305f\u3002\u305d\u306e\u305f\u3081\u3001\u4ee5\u964d\u306e\u8a18\u4e8b\u3067\u306fcpu\u30e2\u30fc\u30c9\u3067\u52d5\u304b\u305b\u308b\u3088\u3046\u306b\u6539\u4fee\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
Python\u30b9\u30af\u30ea\u30d7\u30c8\u3092\u4fee\u6b63<\/h3>\n\n\n\n
\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u52d5\u4f5c\u3055\u305b\u308b\u306b\u3042\u305f\u308a\u3001Python\u30b9\u30af\u30ea\u30d7\u30c8\u306e\u4fee\u6b63\u3092\u884c\u3044\u307e\u3059<\/p>\n\n\n\n
scripts\/txt2img.py<\/strong>
63\u884c\u76ee\u3042\u305f\u308a\u306emodel.cuda()\u306e\u90e8\u5206\u3092\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u4fee\u6b63<\/p>\n\n\n\n
    if len(u) > 0 and verbose:\n        print("unexpected keys:")\n        print(u)\n\n    # model.cuda()\n    model.to('cpu')\n    model.eval()\n    return model<\/code><\/pre><\/div>\n\n\n\n
243\u884c\u76ee\u8fba\u308a\u306edevice = torch.device(“cuda”) if torch.cuda.is_available() else torch.device(“cpu”)\u306e\u90e8\u5206\u3092\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u4fee\u6b63<\/p>\n\n\n\n
    # device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")\n    device = torch.device("cpu")\n    model = model.to(device)<\/code><\/pre><\/div>\n\n\n\n
283\u884c\u76ee\u8fba\u308a\u306ewith precision_scope(“cuda”):\u306e\u90e8\u5206\u3092\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u4fee\u6b63<\/p>\n\n\n\n
    with torch.no_grad():\n        # with precision_scope("cuda"):\n        with precision_scope("cpu"):<\/code><\/pre><\/div>\n\n\n\n
ldm\/models\/diffusion\/plms.py<\/strong>
18\u884c\u76ee\u8fba\u308a\u306eif attr.device != torch.device(“cuda”):\u5468\u8fba\u3092\u4ee5\u4e0b\u306e\u901a\u308a\u4fee\u6b63<\/p>\n\n\n\n
    def register_buffer(self, name, attr):\n        if type(attr) == torch.Tensor:\n            # if attr.device != torch.device("cuda"):\n            #    attr = attr.to(torch.device("cuda"))\n            if attr.device != torch.device("cpu"):\n                attr = attr.to(torch.float32).to(torch.device("cpu")).contiguous()\n        setattr(self, name, attr)<\/code><\/pre><\/div>\n\n\n\n
configs\/stable-diffusion\/v1-inference.yaml<\/strong>
69\u884c\u76ee\u4ed8\u8fd1\u306e\u8a2d\u5b9a\u3092\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u4fee\u6b63<\/p>\n\n\n\n
    cond_stage_config:\n      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder\n      params:\n        device: cpu<\/code><\/pre><\/div>\n\n\n\n
ldm\/modules\/attention.py<\/strong>
211\u884c\u76ee\u4ed8\u8fd1\u306edef _forward(self, x, context=None):\u3092\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u4fee\u6b63<\/p>\n\n\n\n
    def _forward(self, x, context=None):\n        x = x.contiguous()\n        x = self.attn1(self.norm1(x)) + x\n        x = self.attn2(self.norm2(x), context=context) + x\n        x = self.ff(self.norm3(x)) + x\n        return x<\/code><\/pre><\/div>\n\n\n\n
\u3044\u3088\u3044\u3088\u5b9f\u884c\u3057\u3066\u307f\u308b<\/h2>\n\n\n\n
\u4ee5\u4e0a\u306e\u5909\u66f4\u3067\u52d5\u4f5c\u3059\u308b\u3088\u3046\u306b\u306a\u3063\u3066\u3044\u308b\u306f\u305a\u3067\u3059\u3002
\u30c6\u30ad\u30b9\u30c8\u304b\u3089\u30a4\u30e1\u30fc\u30b8\u3092\u4f5c\u6210\u3059\u308b\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u5b9f\u884c\u3057\u3001\u30a4\u30e9\u30b9\u30c8\u3092\u751f\u6210\u3057\u3066\u307f\u307e\u3057\u3087\u3046
(\u53c2\u8003\u30da\u30fc\u30b8\u3068\u540c\u3058\u3060\u3068\u9762\u767d\u304f\u306a\u3044\u306e\u3067\u300c\u4e00\u7dd2\u306b\u5bdd\u3066\u3044\u308b\u72ac\u3068\u732b\u300d\u3068\u3044\u3046\u984c\u76ee\u3067\u30a4\u30e9\u30b9\u30c8\u3092\u751f\u6210\u3057\u307e\u3057\u305f)<\/p>\n\n\n\n
python scripts\/txt2img.py --prompt "cat and dog sleeping together" --plms --precision full<\/code><\/pre><\/div>\n\n\n\n
\u751f\u6210\u3055\u308c\u305f\u753b\u50cf\u306f\u30ea\u30dd\u30b8\u30c8\u30ea\u3092\u30af\u30ed\u30fc\u30f3\u3057\u305f\u76f4\u4e0b\u306e\u30d5\u30a9\u30eb\u30c0\u914d\u4e0b\u306b\u51fa\u529b\u3055\u308c\u307e\u3059
outputs\/txt2img-samples\/<\/p>\n\n\n\n
\u30a4\u30e9\u30b9\u30c8\u304c\u751f\u6210\u3055\u308c\u308b\u307e\u30671\u6642\u959340\u5206\u307b\u3069\u304b\u304b\u308a\u307e\u3057\u305f\u3002\u751f\u6210\u3055\u308c\u305f\u753b\u50cf\u306f\u4ee5\u4e0b\u306b\u306a\u308a\u307e\u3059<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\u3056\u3063\u3068\u898b\u305d\u308c\u3063\u307d\u3044\u753b\u50cf\u304c\u751f\u6210\u3055\u308c\u3066\u3044\u307e\u3059\u304c\u3001\u3088\u304f\u898b\u308b\u3068\u5f62\u304c\u3044\u3073\u3064\u306a\u90e8\u5206\u304c\u3042\u3063\u305f\u308a\u3068\u3061\u3087\u3063\u3068\u6016\u3044\u30fb\u30fb\u30fb
\u305f\u3060\u3001\u3053\u306e\u30af\u30aa\u30ea\u30c6\u30a3\u306e\u753b\u50cf\u304c2\u6642\u9593\u4ee5\u5185\u3067\u751f\u6210\u3055\u308c\u308b\u3068\u3044\u3046\u306e\u306f\u3001\u3059\u3054\u3044\u3053\u3068\u306a\u306e\u3067\u306f\u3068\u601d\u3044\u307e\u3059<\/p>\n\n\n\n
\u540c\u3058\u3088\u3046\u306a\u30b5\u30fc\u30d3\u30b9\u3067\u300cmidjourney\u300d\u304c\u6709\u540d\u3067\u3059\u304c\u3001\u3042\u3061\u3089\u306f\u6570\u5206\u3067\u753b\u50cf\u304c\u751f\u6210\u3055\u308c\u308b\u306e\u3067\u304b\u306a\u308a\u5f37\u529b\u306a\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u3067\u51e6\u7406\u3057\u3066\u308b\u3093\u3060\u308d\u3046\u306a\u3068\u601d\u3044\u307e\u3059\u3002<\/p>\n\n\n\n
\u4ed6\u306b\u3082\u3001\u6700\u8fd1\u300cmimic<\/a>\u300d\u3068\u3044\u3046\u30a4\u30e9\u30b9\u30c8\u751f\u6210\u30b5\u30fc\u30d3\u30b9\u304c\u30ea\u30ea\u30fc\u30b9\u3055\u308c\u307e\u3057\u305f\u304c\u3001\u3053\u3061\u3089\u306f\u8272\u3005\u7269\u8b70\u3092\u91b8\u3057\u3066\u3044\u308b\u3088\u3046\u3067\u3059\u3002
\u4eca\u5f8c\u3082\u30a4\u30e9\u30b9\u30c8AI\u306e\u9032\u5316\u306f\u7d9a\u304f\u3068\u601d\u3044\u307e\u3059\u304c\u3001\u30af\u30ea\u30a8\u30a4\u30bf\u30fc\u3068AI\u304c\u3046\u307e\u304f\u5171\u5b58\u3067\u304d\u308b\u3088\u3046\u306a\u30b5\u30fc\u30d3\u30b9\u304c\u8a95\u751f\u3059\u308b\u3053\u3068\u3092\u9858\u3063\u3066\u3044\u307e\u3059\u3002
<\/p>\n\n\n\n
\u4eca\u56de\u306f\u3053\u3053\u307e\u3067<\/h2>\n\n\n\n
\u3072\u3068\u307e\u305a\u7121\u4e8b\u306b\u30a4\u30e9\u30b9\u30c8\u751f\u6210\u304c\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u3063\u305f\u3068\u3053\u308d\u3067\u3001\u4eca\u56de\u306f\u4e00\u65e6\u7d42\u4e86\u3002
\u30b9\u30af\u30ea\u30d7\u30c8\u30d5\u30a9\u30eb\u30c0\u306e\u4e2d\u306b\u3001img2img\u3084knn2img\u306a\u3069\u9762\u767d\u305d\u3046\u306a\u30b9\u30af\u30ea\u30d7\u30c8\u304c\u6e96\u5099\u3055\u308c\u3066\u3044\u305f\u306e\u3067\u3001\u3053\u3061\u3089\u306f\u307e\u305f\u6642\u9593\u3092\u898b\u3064\u3051\u3066\u8a66\u3057\u3066\u307f\u305f\u3044\u3068\u601d\u3044\u307e\u3059<\/p>\n\n\n\n
\u30a8\u30e9\u30fc\u96c6<\/h2>\n\n\n\n
Traceback (most recent call last):\n  File "scripts\/txt2img.py", line 347, in <module>\n    main()\n  File "scripts\/txt2img.py", line 285, in main\n    with precision_scope("mps"):\n  File "\/opt\/anaconda3\/envs\/ldm\/lib\/python3.8\/site-packages\/torch\/amp\/autocast_mode.py", line 199, in __init__\n    raise RuntimeError('User specified autocast device_type must be \\'cuda\\' or \\'cpu\\'')\nRuntimeError: User specified autocast device_type must be 'cuda' or 'cpu'<\/code><\/pre><\/div>\n\n\n\n
\u30d7\u30ed\u30b0\u30e9\u30e0\u4fee\u6b63\u9593\u9055\u3044\u304c\u3042\u308a\u767a\u751f\u3002\u898b\u76f4\u3057\u3066\u89e3\u6d88<\/p>\n\n\n\n
RuntimeError: Invalid buffer size: 3.00 GB<\/code><\/pre><\/div>\n\n\n\n
\u81ea\u5206\u306e\u74b0\u5883\u3067\u306fmps\u3092\u5229\u7528\u3059\u308b\u3068\u4e0a\u8a18\u30a8\u30e9\u30fc\u304c\u767a\u751f\u3057\u3066\u3057\u307e\u3044\u307e\u3057\u305f\u3002
cpu\u3092\u5229\u7528\u3059\u308b\u30e2\u30fc\u30c9\u306b\u5909\u66f4\u3057\u3066\u89e3\u6c7a\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n
RuntimeError: expected scalar type BFloat16 but found Float<\/code><\/pre><\/div>\n\n\n\n
\u4ee5\u4e0b\u306e\u8a18\u4e8b\u3092\u53c2\u8003
https:\/\/zenn.dev\/isaka102\/scraps\/2a37006ed35248<\/a><\/p>\n\n\n\n
\u5f15\u6570\u306b\u300c–precision full\u300d\u3092\u8ffd\u52a0\u3057\u3066\u89e3\u6c7a<\/p>\n\n\n
\n\t
\n\t\t
\n\t\t\t\t\t\t\t<\/a>\t\t\t\t\t<\/div>\n\t\t
\n\t\t\t
\n\t\t\t\t\t\t\t\t\t\u300c\u30ad\u30e3\u30e9\u30af\u30bf\u30fc\u300d\u306e\u30c7\u30b6\u30a4\u30f3\uff06\u63cf\u304d\u65b9\u3000\u30ab\u30e9\u30d5\u30eb\u30dd\u30c3\u30d7\u3067\u9b45\u305b\u308b\u30a4\u30e9\u30b9\u30c8\u6280\u5de7 \uff08\u30db\u30d3\u30fc\u30b8\u30e3\u30d1\u30f3\u306e\u6280\u6cd5\u66f8\uff09 [ \u304f\u308b\u307f\u3064 ]<\/a>\t\t\t\t\t\t\t<\/div>\n\t\t\t
\n\t\t\t\t\t\t\t
created by Rinker<\/a><\/div>\n\t\t\t\t\t\t\t\t\t\t
\n\t\t\t\t\t\t\t\u00a52,420<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t(2026\/04\/28 06:31:33\u6642\u70b9 \u697d\u5929\u5e02\u5834\u8abf\u3079-<\/span>\u8a73\u7d30)<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t