Unable to generate index file

[burnt basin]

Translated into English: After completing training with the v4 image provided by “rvc-boss” in autodl, an error message appeared stating: “UserWarning: resource_tracker: There appear to be 20 leaked semaphore objects to clean up at shutdown. warnings.warn('resource_tracker: There appear to be %d '”
To resolve this issue, I have tried: changing GPUs (4090-3080-A100-A4000), switching to the 1006 image, reducing the training set (from 1 hour to 30 minutes, 10 minutes), and decreasing the total_epoch (from 1000 to 200, 20, 10), but none of these solutions worked. Even training a 5-minute training set for 10 rounds resulted in the same error message after training completion, and then there was no index file. When I manually click on the training index, the web indicates training but then there is no response.

使用autodl中的“rvc-boss”提供的v4镜像训练完成后出现错误提示“UserWarning: resource_tracker: There appear to be 20 leaked semaphore objects to clean up at shutdown
warnings.warn('resource_tracker: There appear to be %d '”
为了解决这个问题，我尝试过：更换gpu（4090-3080-A100-A4000），更换1006镜像，减少训练集（从1小时减少到30分钟，10分钟），减少训练轮数（从1000减少到200，20，10），都未能解决,甚至在5分钟的训练集训练10轮，都是在训练完成后出现这个错误提示然后就没有index文件，当我手动点击训练索引时web提示training然后没有任何反应

[stable compass]

I have the same issue: #🏋訓練training message

[ornate wave]

@stable compass has reached level 1. GG!

[lean gale]

it looks like it built successfully. training the index file can just take a long time. that "leaked semaphore" message always appears after the training hits its end goal

[burnt basin]

I waited for 30 minutes, but no index appeared.

[lean gale]

how big was your dataset?

length-wise

[burnt basin]

59min

[ornate wave]

@burnt basin has reached level 1. GG!

[burnt basin]

Even if I reduce the dataset to 10 minutes, it still gives this prompt and does not generate an index

[lean gale]

try generating the index first maybe. it doesn't matter when you generate it. the more audio you have the longer it'll take to generate and the bigger it'll be

[burnt basin]

Generate index directly without training model?

[lean gale]

correct. the index can be generated after feature extraction. the model doesn't have to train to generate an index

[burnt basin]

Click on the training index and this appears. No index file is generated.

[lean gale]

that "training" means it's training the index file

give it time, eventually it should say "added"

[burnt basin]

ok,thx

[stable compass]

Interesting. Any luck @burnt basin? Trying this now, I'm not seeing any significant CPU usage or "Energy" usage (macOS "Activity Monitor" app). I think I found the Python process running inside bash inside zsh, or at least one of them, based on the list of open files, but it's only using (on average) about 1.7% CPU "Energy". Is that normal?

[ornate wave]

@stable compass has reached level 2. GG!

[stable compass]

I'm currently trying to generate an index from only 27 seconds of audio, to confirm it's possible with my setup (M1 MBP), but I'm hoping to use closer to 30 minutes of audio if that succeeds.

[stable compass]

Running for over 6 hours, the Python process has read over 4.5 GB of data (but only written 3.3 MB), it's still reading. Again, from only 27 seconds of audio. Minimal CPU usage, 0.6%.