关于中文特化版本train_ms.py训练问题 #298
Unanswered
windforestfiremountain
asked this question in
Q&A
Replies: 4 comments 1 reply
-
conda list 的环境如下:
|
Beta Was this translation helpful? Give feedback.
1 reply
-
您好,我遇到了相同问题,请问你解决了吗? |
Beta Was this translation helpful? Give feedback.
0 replies
-
@windforestfiremountain @andymethane 项目代码用的切片方法slice_segments和原始VITS的有差别,所以不能正常地在python中报错,而是报了底层c++的错误。改成下面的函数即可正常运行。 commons.py - slice_segments def slice_segments(x, ids_str, segment_size=4):
ret = torch.zeros_like(x[:, :, :segment_size])
for i in range(x.size(0)):
idx_str = ids_str[i]
idx_end = idx_str + segment_size
if idx_end > x.size(-1):
# pad x
x = F.pad(x, (0, (idx_end - x.size(-1)), 0, 0), "constant", 0)
try:
ret[i] = x[i, :, idx_str:idx_end]
except Exception as e:
print(i, x.shape, idx_str, idx_end)
raise e
return ret |
Beta Was this translation helpful? Give feedback.
0 replies
-
检查下n_speakers是否正确 |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
作者您好,在使用最新中文特化版本调用的时候出现了分布式训练报错(硬件环境4090四卡config.yml配置WORLD_SIZE: 1
,LOCAL_RANK: 1,RANK: 12),报错日志如下,请问有什么解决方法吗?
Beta Was this translation helpful? Give feedback.
All reactions