Does moss-tts-realtime support concurrent requests on a single instance?

[richumble]

I’m trying to handle 5 concurrent clients using one model instance. However, I noticed voice distortion and heavy CPU spikes when requests run in parallel.

Is the run_stream logic thread-safe?

How should I handle multiple concurrent tasks without quality degradation?

[YWMditto]

We will be open-sourcing a thread-safe version of sglang soon, so please stay tuned.