【2026.04.30更新】Deepfacelab 猫之汉化pytorch版

wtxx8888

675966766 发表于 2026-5-9 23:08
我试几次把增量包解压覆盖进去，包括整个重新下载都没有BF16和真脸强度开关 ...

BF16最低   要30系显卡。

gaolaodao7788

大佬啊，用这个版本跑模型速度非常快，导出SAEHD模型的时候就出问题啦。这个怎么调请指教。拜托
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

正在导出 .dfm 到 D:\MAO-DeepFaceLab-Pytorch\workspace\直播模型\256-WF-D_SAEHD.dfm
[torch.onnx] Obtain model graph for `DFMModel()` with `torch.export.export(..., strict=False)`...
[torch.onnx] Obtain model graph for `DFMModel()` with `torch.export.export(..., strict=False)`... ❌
[torch.onnx] Obtain model graph for `DFMModel()` with `torch.export.export(..., strict=True)`...
[torch.onnx] Obtain model graph for `DFMModel()` with `torch.export.export(..., strict=True)`... ❌
/!\ 导出失败: Failed to export the model with torch.export. This is step 1/3 of exporting the model to ONNX. Next steps:
- Modify the model code for `torch.export.export` to succeed. Refer to https://pytor
Traceback (most recent call last):
  File "D:\MAO-DeepFaceLab-Pytorch\_internal\python\Lib\site-packages\torch\onnx\_internal\exporter\_capture_strategies.py", line 121, in __call__
    exported_program = self._capture(model, args, kwargs, dynamic_shapes)
  File "D:\MAO-DeepFaceLab-Pytorch\_internal\python\Lib\site-packages\torch\onnx\_internal\exporter\_capture_strategies.py", line 219, in _capture
    return torch.export.export(
  File "D:\MAO-DeepFaceLab-Pytorch\_internal\python\Lib\site-packages\torch\export\__init__.py", line 311, in export

......
......
<class 'RuntimeError'>: shape '[1, 2, 2, 0, 128, 512]' is invalid for input of size 196608

zhuntiancair

猫之汉化 发表于 2026-5-1 16:56
把增量更新包放进去覆盖，把bf16和梯度检查打开

大佬怎么使用这个增量包啊，我直接覆盖到DeepFaceLab-Pytorch文件夹里吗，我用的4070显卡，不知道运行哪个文件了，求指教

zhuntiancair

大佬，如何开半精度啊，没找到开关，求指教

sex9868

增量更新覆盖后就会出错，没更新前是好的，========================================
检测到多个可用的GPU后端:
  1 - CUDA (NVIDIA显卡)
  2 - DirectML (AMD/Intel显卡)
========================================
请输入 [1/2] (回车=CUDA):
Traceback (most recent call last):
  File "<frozen __main__>", line 3, in <module>
  File "<frozen main>", line 7, in <module>
AttributeError: module 'core.leras.nn' has no attribute 'initialize_main_env'
Press any key to continue . . .

iou99vip

zhuntiancair 发表于 2026-5-15 15:49
大佬怎么使用这个增量包啊，我直接覆盖到DeepFaceLab-Pytorch文件夹里吗，我用的4070显卡，不知道运行哪 ...

复制到_internal\DeepFaceLab

dw123

合成稍微有点慢，是做了优化吗？感觉对比原版的和神农版的同配置同参数下慢了几倍

ajno22

合成阶段到出现需要换脸的那一帧的时候出现错误，好像读取不到了然后错误，这个问题是什么情况啊，第一次遇见

Traceback (most recent call last):
  File "<frozen DeepFaceLab.mainscripts.Merger>", line 209, in main
  File "<frozen SubprocessorBase>", line 266, in run
  File "<frozen interact>", line 186, in process_messages
  File "<frozen DeepFaceLab.core.joblib.MPClassFuncOnDemand>", line 22, in io_callback
  File "<frozen XSegNet>", line 80, in __init__
  File "<frozen DeepFaceLab.core.leras.models.ModelBase>", line 94, in build
TypeError: XSeg.on_build() got an unexpected keyword argument 'resolution'

dyhanming

file:///E:/Mydata/Pictures/DFL%E6%8A%A5%E9%94%99.png
训练了2天，报错，重新拷贝软件运行，还是报错，怎么解决？

zslj

开启BF16情况下训练出现时快时慢的情况，一次快一次慢，看了下显卡占用是断断续续的，有人知道原因吗，显卡是5060ti16g，最新的stuido驱动