新手求教：liae结构的迪丽热巴万能模型 训练问题

km123

星级打分

• 1
• 2
• 3
• 4
• 5

平均分:NAN  参与人数:0  我的评分:未评

能直接合成，但换其它src训练时会出错：大神们帮看下，谢谢
保存时间|迭代次数|单次时间|SRC损失|DST损失
Error: 2 root error(s) found.
  (0) Resource exhausted: OOM when allocating tensor with shape[3,3,2048,2048] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
         [[node src_dst_opt_1/Select_26 (defined at H:\DFL_maozhihanhua_RTX3000\_internal\DeepFaceLab\core\leras\ops\__init__.py:212) ]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info. This isn't available when running in Eager mode.

         [[concat_7/concat/_963]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info. This isn't available when running in Eager mode.

  (1) Resource exhausted: OOM when allocating tensor with shape[3,3,2048,2048] and type float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator GPU_0_bfc
         [[node src_dst_opt_1/Select_26 (defined at H:\DFL_maozhihanhua_RTX3000\_internal\DeepFaceLab\core\leras\ops\__init__.py:212) ]]
Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info. This isn't available when running in Eager mode.

0 successful operations.
0 derived errors ignored.

Errors may have originated from an input operation.
Input Source operations connected to node src_dst_opt_1/Select_26:
src_dst_opt_1/Less_26 (defined at H:\DFL_maozhihanhua_RTX3000\_internal\DeepFaceLab\core\leras\ops\__init__.py:211)

Input Source operations connected to node src_dst_opt_1/Select_26:
src_dst_opt_1/Less_26 (defined at H:\DFL_maozhihanhua_RTX3000\_internal\DeepFaceLab\core\leras\ops\__init__.py:211)

Original stack trace for 'src_dst_opt_1/Select_26':
  File "threading.py", line 884, in _bootstrap
  File "threading.py", line 916, in _bootstrap_inner
  File "threading.py", line 864, in run
  File "H:\DFL_maozhihanhua_RTX3000\_internal\DeepFaceLab\mainscripts\Trainer.py", line 58, in trainerThread
    debug=debug)
  File "H:\DFL_maozhihanhua_RTX3000\_internal\DeepFaceLab\models\ModelBase.py", line 199, in __init__
    self.on_initialize()
  File "H:\DFL_maozhihanhua_RTX3000\_internal\DeepFaceLab\models\Model_SAEHD\Model.py", line 341, in on_initialize

ycjack27

hideto123 发表于 2022-7-10 08:08
那么请问下470W-带GAN和 300W-RW开 新手选哪个好出效果呢？src素材不太高清，3060 12g显卡 ...

区别不大，反正练新的src都是要开RW的，带GAN的开了RW跑一会GAN的效果就没了

20210901

显存太低， 升级大显存显卡

hideto123

20210901 发表于 2022-7-7 17:57
显存太低， 升级大显存显卡

要换其他src怎么操作？在文件夹里把inter ab删了就可以为什么删了第二第屋列的图片就没了？

ycjack27

hideto123 发表于 2022-7-8 15:29
要换其他src怎么操作？在文件夹里把inter ab删了就可以为什么删了第二第屋列的图片就没了？ ...

可以不用删AB，效果差不多，都是两三万次迭代就出效果了。删了也一样，跑一会就行

ycjack27

显存不够，但这个万能模型配置还好，一般的30系显卡应该都能跑的，你是不是20系的，然后用了这个30系版本的dfl

hideto123

ycjack27 发表于 2022-7-9 10:29
可以不用删AB，效果差不多，都是两三万次迭代就出效果了。删了也一样，跑一会就行 ...

那么请问下470W-带GAN和 300W-RW开 新手选哪个好出效果呢？src素材不太高清，3060 12g显卡

hideto123

ycjack27 发表于 2022-7-11 09:57
区别不大，反正练新的src都是要开RW的，带GAN的开了RW跑一会GAN的效果就没了 ...

RW是随即扭曲吗？意思是我用这个模型训练前要按2下回车那个重新设置参数？

ycjack27

hideto123 发表于 2022-7-11 12:03
RW是随即扭曲吗？意思是我用这个模型训练前要按2下回车那个重新设置参数？ ...

对，否则不像src