CUDA out of memory. Tried to allocate 1.71 GiB (GPU 0; 11.00 GiB total capacity; 8.78 GiB already allocated; 0 bytes free; 9.92 GiB reserved in total by PyTorch) #548
Unanswered
a-cold-bird
asked this question in
Q&A
Replies: 2 comments 1 reply
-
突然发现我用其他的歌可以正常转换?但是爆显存的歌时长只有2分钟,原本3,4分钟的歌都可以推理来着,也没用到切片。请问一下什么因素会导致推理的时间变长或者导致显存占用更多呢 |
Beta Was this translation helpful? Give feedback.
1 reply
-
既然是爆显存的问题,那就把歌切割成更小块就可以了 爆显存跟歌的内容也有关系的,不单单只看时长的 还有你用的是整合包吗,如果是整合包的话,可能会有些奇怪的bug的 |
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
-
这是在由svc-develop-svc的项目so vits上出现的问题,由于无法在其仓库提交issue就到这问问了,很抱歉打扰到各位
load
INFO:root:Loaded checkpoint 'D:\my_models\vits4_models\renge\renge_28800.pth' (iteration 375)
#=====segment start, 123.54s======
Traceback (most recent call last):
File "inference_main.py", line 104, in
main()
File "inference_main.py", line 88, in main
out_audio, out_sr = svc_model.infer(spk, tran, raw_path,
File "D:\BaiduNetdiskDownload\So-VITS-SVC\sovits\so-vits-svc-4.0\inference\infer_tool.py", line 177, in infer
audio = self.net_g_ms.infer(c, f0=f0, g=sid, uv=uv, predict_f0=auto_predict_f0, noice_scale=noice_scale)[0,0].data.float()
File "D:\BaiduNetdiskDownload\So-VITS-SVC\sovits\so-vits-svc-4.0\models.py", line 417, in infer
z_p, m_p, logs_p, c_mask = self.enc_p(x, x_mask, f0=f0_to_coarse(f0), noice_scale=noice_scale)
File "D:\so-vits-svc\Dependencies\lib\site-packages\torch\nn\modules\module.py", line 1194, in call_impl
return forward_call(*input, **kwargs)
File "D:\BaiduNetdiskDownload\So-VITS-SVC\sovits\so-vits-svc-4.0\models.py", line 114, in forward
x = self.enc(x * x_mask, x_mask)
File "D:\so-vits-svc\Dependencies\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "D:\BaiduNetdiskDownload\So-VITS-SVC\sovits\so-vits-svc-4.0\modules\attentions.py", line 85, in forward
y = self.attn_layers[i](x, x, attn_mask)
File "D:\so-vits-svc\Dependencies\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "D:\BaiduNetdiskDownload\So-VITS-SVC\sovits\so-vits-svc-4.0\modules\attentions.py", line 189, in forward
x, self.attn = self.attention(q, k, v, mask=attn_mask)
File "D:\BaiduNetdiskDownload\So-VITS-SVC\sovits\so-vits-svc-4.0\modules\attentions.py", line 221, in attention
relative_weights = self._absolute_position_to_relative_position(p_attn)
File "D:\BaiduNetdiskDownload\So-VITS-SVC\sovits\so-vits-svc-4.0\modules\attentions.py", line 287, in _absolute_position_to_relative_position
x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.71 GiB (GPU 0; 11.00 GiB total capacity; 8.78 GiB already allocated; 0 bytes free; 9.92 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
在之前程序是能正常运行的,11g显存仅仅只是用于推理,也没有开启f0滤波器,但是还是不知道为什么显存爆了,目前已经试过了最新版本的so vits以及旧版本不带f0滤波器的版本,模型也都换过,都会在“segment start, 123.54s“显示后不久报错,从任务管理器返回的显存占用来看,显存占用最开始正常3g左右,过一会就直接吃满,过一会就跳cuda out of memory。请问一下应该怎么解决这个问题,我在其他报错解决方案后尝试在inference.py中导入os之后限制调用内存还是会报这个错误
使用window10
python3.8
显卡2080ti 11g
请问一下应该怎么解决这个问题,我在其他报错解决方案后尝试在inference.py中导入os之后限制调用内存还是会报这个错误
不胜感激
Beta Was this translation helpful? Give feedback.
All reactions