Mark Robert Miller

I've put an initial before/after comparison for JavaBinCodec decode performance in the JIRA issue: https://issues.apache.org/jira/browse/SOLR-15560

So far, initial summary is: 79% improvement for decode performance for this benchmark. 327% improvement for encode performance for this benchmark.

Review on this WIP code from 29 days has little value.

Onthe RVC website, it says the training data is an English speaking dataset, as far as I've seen in these sovits 5 projects it's a singing dataset and I believe...

> Some developers reported hallucination problem of large v3, hope there will be a comparison to prove whether there's improvement or not. [openai/whisper#1762 (comment)](https://github.com/openai/whisper/discussions/1762#discussioncomment-7557044) Some people much prefer V2, but...

Interesting. On my attempt I got the same type of samples but also go then with the minimal infer.

Maybe the issue I had was due to using Adam 8-bit. Missed the note it didn't work. So I changed to recommend. Model wouldn't train. Found the notebook for this...