请保留GPU选择，即使只有一个GPU

[lo-bei-ve]

我之前是双卡，使用的GPU1训练，但是最近我换成单卡了，训练参数还是默认的GPU1，并且在参数配置界面无法切换
21:16:26-798101 INFO Training started with config file / 训练开始，使用配置文件:
C:\Users\wkl\Desktop\lora-scripts-v1.10.0\config\autosave\20251018-211626.toml
21:16:26-800311 INFO Using GPU(s) / 使用 GPU: ['1']
21:16:26-801939 INFO Task 2e7044f7-be93-472b-b988-956f8c24d4b4 created
2025-10-18 21:16:34 INFO Loading settings from train_util.py:4012
C:\Users\wkl\Desktop\lora-scripts-v1.10.0\config\autosave\20251018-21162
6.toml...
INFO C:\Users\wkl\Desktop\lora-scripts-v1.10.0\config\autosave\20251018-21162 train_util.py:4031
6
2025-10-18 21:16:34 INFO prepare tokenizer train_util.py:4558
21:16:35-870893 ERROR Training failed / 训练失败

[lo-bei-ve]

20:54:04-406143 INFO Training started with config file / 训练开始，使用配置文件:
C:\Users\wkl\Desktop\lora-scripts-v1.10.0\config\autosave\20251018-205404.toml
20:54:04-407647 INFO Using GPU(s) / 使用 GPU: ['1']
20:54:04-411154 INFO Task c8e1d65c-0119-4842-8058-d4eccf3671e5 created
2025-10-18 20:54:12 INFO Loading settings from train_util.py:4012
C:\Users\wkl\Desktop\lora-scripts-v1.10.0\config\autosave\20251018-20540
4.toml...
INFO C:\Users\wkl\Desktop\lora-scripts-v1.10.0\config\autosave\20251018-20540 train_util.py:4031
4
2025-10-18 20:54:12 INFO prepare tokenizers sdxl_train_util.py:138
2025-10-18 20:54:15 INFO update token length: 255 sdxl_train_util.py:163
INFO Using DreamBooth method. train_network.py:172
WARNING max_bucket_reso is adjusted to be multiple of bucket_reso_steps / train_util.py:669
max_bucket_resoがbucket_reso_stepsの倍数になるように調整されました: 1560
-> 1600
INFO prepare images. train_util.py:1686
INFO found directory C:\Users\wkl\Desktop\lora\lj5\5_liujie contains 112 train_util.py:1633
image files
INFO 560 train images with repeating. train_util.py:1727
INFO 0 reg images. train_util.py:1730
WARNING no regularization images / 正則化画像が見つかりませんでした train_util.py:1735
INFO [Dataset 0] config_util.py:572
batch_size: 10
resolution: (1280, 960)
enable_bucket: True
network_multiplier: 1.0
min_bucket_reso: 256
max_bucket_reso: 1600
bucket_reso_steps: 64
bucket_no_upscale: True

                           [Subset 0 of Dataset 0]
                             image_dir: "C:\Users\wkl\Desktop\lora\lj5\5_liujie"
                             image_count: 112
                             num_repeats: 5
                             shuffle_caption: True
                             keep_tokens: 1
                             keep_tokens_separator:
                             caption_separator: ,
                             secondary_separator: None
                             enable_wildcard: False
                             caption_dropout_rate: 0.05
                             caption_dropout_every_n_epoches: 0
                             caption_tag_dropout_rate: 0.05
                             caption_prefix: None
                             caption_suffix: None
                             color_aug: False
                             flip_aug: False
                             face_crop_aug_range: None
                             random_crop: False
                             token_warmup_min: 1,
                             token_warmup_step: 0,
                             alpha_mask: False,
                             is_reg: False
                             class_tokens: liujie
                             caption_extension: .txt


                INFO     [Dataset 0]                                                              config_util.py:578
                INFO     loading image sizes.                                                      train_util.py:901

100%|██████████████████████████████████████████████████████████████████████████████| 112/112 [00:00<00:00, 1773.36it/s]
INFO make buckets train_util.py:907
WARNING min_bucket_reso and max_bucket_reso are ignored if bucket_no_upscale is train_util.py:924
set, because bucket reso is defined by image size automatically /
bucket_no_upscaleが指定された場合は、bucketの解像度は画像サイズから自動計
算されるため、min_bucket_resoとmax_bucket_resoは無視されます
INFO number of images (including repeats) / train_util.py:953
各bucketの画像枚数（繰り返し回数を含む）
INFO bucket 0: resolution (960, 1280), count: 40 train_util.py:958
INFO bucket 1: resolution (1088, 1088), count: 25 train_util.py:958
INFO bucket 2: resolution (1280, 960), count: 495 train_util.py:958
INFO mean ar error (without repeats): 0.0 train_util.py:963
INFO preparing accelerator train_network.py:225
accelerator device: cpu
INFO loading model for process 0/1 sdxl_train_util.py:33
INFO load StableDiffusion checkpoint: sdxl_train_util.py:74
D:/ComfyUI-aki-v1.6/ComfyUI/models/checkpoints/ultra_v10.safetensors
2025-10-18 20:54:16 INFO building U-Net sdxl_model_util.py:198
INFO loading U-Net from checkpoint sdxl_model_util.py:202
2025-10-18 20:54:30 INFO U-Net: sdxl_model_util.py:208
INFO building text encoders sdxl_model_util.py:211
INFO loading text encoders from checkpoint sdxl_model_util.py:264
2025-10-18 20:54:31 INFO text encoder 1: sdxl_model_util.py:278
2025-10-18 20:54:35 INFO text encoder 2: sdxl_model_util.py:282
INFO building VAE sdxl_model_util.py:285
INFO loading VAE from checkpoint sdxl_model_util.py:290
2025-10-18 20:54:36 INFO VAE: sdxl_model_util.py:293
INFO Enable xformers for U-Net train_util.py:2832
Traceback (most recent call last):
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\scripts\stable\sdxl_train_network.py", line 185, in
trainer.train(args)
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\scripts\stable\train_network.py", line 242, in train
vae.set_use_memory_efficient_attention_xformers(args.xformers)
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\python\lib\site-packages\diffusers\models\modeling_utils.py", line 262, in set_use_memory_efficient_attention_xformers
fn_recursive_set_mem_eff(module)
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\python\lib\site-packages\diffusers\models\modeling_utils.py", line 258, in fn_recursive_set_mem_eff
fn_recursive_set_mem_eff(child)
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\python\lib\site-packages\diffusers\models\modeling_utils.py", line 258, in fn_recursive_set_mem_eff
fn_recursive_set_mem_eff(child)
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\python\lib\site-packages\diffusers\models\modeling_utils.py", line 258, in fn_recursive_set_mem_eff
fn_recursive_set_mem_eff(child)
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\python\lib\site-packages\diffusers\models\modeling_utils.py", line 255, in fn_recursive_set_mem_eff
module.set_use_memory_efficient_attention_xformers(valid, attention_op)
File "C:\Users\wkl\Desktop\lora-scripts-v1.10.0\python\lib\site-packages\diffusers\models\attention_processor.py", line 260, in set_use_memory_efficient_attention_xformers
raise ValueError(
ValueError: torch.cuda.is_available() should be True but is False. xformers' memory efficient attention is only available for GPU
20:54:37-563648 ERROR Training failed / 训练失败

这里有更详细的报错信息，我到现在还没找到配置文件储存在哪里，头疼

[Ice-YY]

看上去这个是配置文件？
C:\Users\wkl\Desktop\lora-scripts-v1.10.0\config\autosave\20251018-21162
6.toml
具体可以参考https://github.com/kohya-ss/sd-scripts/blob/sd3/docs/sdxl_train_network.md

不过我看了下你的日志，你是不是应该去你用的lora-scripts那边问？https://github.com/Akegarasu/lora-scripts