Ruoyi

icon indicating a website link https://potato-mika.github.io/

icon company 39 AI Inc. icon location Nanjing icon location Time to do something that can change the world.

Results 71 comments of Ruoyi

how to use vllm or trt improve inference?

Not supported in open-source repo.

[Bug]After fine-tuning, the sentence will end only after the first sentence is correct and there are many wrong words afterwards.

请根据自己的数据集大小调整微调步数。

[BUG]字数较少时，生成的音频内容存在问题

短句可能出现幻觉，不建议生成过短音频。

[BUG]The API call does not seem to support wav and pcm.

Closed as the issue has been solved.

[BUG] How many epoch should we finetune with lora?

Fit with the size of your dataset.

Load faster whisper model

自行修改tools/auto_rerank.py中的内容。

The numbers in the text of English to speech conversion have always been in Chinese pronunciation

Any examples?

The numbers in the text of English to speech conversion have always been in Chinese pronunciation

We suggest you to use words instead of arabic numbers.

The numbers in the text of English to speech conversion have always been in Chinese pronunciation

> > use "normalize": false in your api params > > What does `"normalize": false` do? convert arabic numbers into chinese words.

请问微调需要多少数据量？

建议10小时起步