Streaming inference much slower

[canonhui]

i use post_api.py to do inference, and for the exactly same reference audio and to-be-generated-text, when i don't pass the streaming argument, the inference takes like 2 seconds; but if i pass the streaming argument, the inference time increases to like 8 seconds! what could be the reason?

[AnyaCoder]

Maybe it triggered recompilation. Try to update code to latest commit.