50XX,cu12.8适配

[pptbox]

基础 环境：
(Wan21) D:\ai\Wan2.1\flash-attention>nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Fri_Feb_21_20:42:46_Pacific_Standard_Time_2025
Cuda compilation tools, release 12.8, V12.8.93
Build cuda_12.8.r12.8/compiler.35583870_0

(Wan21) D:\ai\Wan2.1\flash-attention>nvidia-smi
Tue Apr 15 08:50:01 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 572.70 Driver Version: 572.70 CUDA Version: 12.8 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Driver-Model | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 5070 Ti WDDM | 00000000:01:00.0 On | N/A |
| 31% 31C P8 22W / 300W | 182MiB / 16303MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 6840 C+G C:\Windows\explorer.exe N/A |
| 0 N/A N/A 8072 C+G ...h_cw5n1h2txyewy\SearchApp.exe N/A |
| 0 N/A N/A 8420 C+G ...xyewy\ShellExperienceHost.exe N/A |
| 0 N/A N/A 10660 C+G ...5n1h2txyewy\TextInputHost.exe N/A |
+-----------------------------------------------------------------------------------------+

#解决5070ti底下不能运行的几个关键点：
1、修改requirements.txt文件
注释：
torch>=2.0.1,<=2.2.2
torchvision>=0.15.2,<=0.17.2
torchaudio>=2.0.2,<=2.2.2

修改：
pip istall diffusers==0.31.0

2、修改EchoMimic\src\models下的这几个文件：
transformer_2d.py
将第7行的：from diffusers.models.embeddings import CaptionProjection
修改为：from diffusers.models.embeddings import PixArtAlphaTextProjection as CaptionProjection

unet_2d_blocks.py
将第9行的：from diffusers.models.dual_transformer_2d import DualTransformer2DModel
修改为：from diffusers.models.transformers.dual_transformer_2d import DualTransformer2DModel

unet_2d_condition.py
将第23行的： PositionNet,
修改为： GLIGENTextBoundingboxProjection as PositionNet,

然后命令行推理即可使用，标准推理5秒大概为6分02秒，
使用acc后推理5秒音频大概为30秒。