Run Higgs without SGlang
I have installed HiggsAudio v3 from the Pinokio app. However, nothing happens. It says “Run the SGLang-Omni backend first”. Do I really have to run SGLang to run this model?
Hey @Syams , you can also use our API for try our model. It's currently free. https://docs.boson.ai/models/higgs-audio-tts/overview
Does this mean that Higgs Audio v3 can’t work completely offline because it requires a web API?
it requires web API? no local offline - is that a joke?? tf
The model is fully open-source and supports 100% local, offline deployment.
The API was only suggested as a free, alternative option for a quick trial. You do not need a web API to run it locally. Please check our blog for local deployment details: https://www.boson.ai/blog/higgs-audio-v3-tts
"API was only suggested as a free, alternative option for a quick trial" why would that even be suggested if it could be run fully locally? I'm having problems running it locally without SGLang-Omni backend. Can you make your instruction clearer for full offline deployment please?
Why suggest the API?
Because setting up a local inference engine (like SGLang) can be time-consuming. We offered the free API purely as a courtesy so users could hear the audio quality immediately before spending time configuring their local environment.Regarding SGLang-Omni:
SGLang-Omni IS a local, 100% offline backend. Running it "locally" means you run the SGLang server on your own machine/GPU—it does not communicate with the internet or any web APIs.
To help us troubleshoot your local deployment issues, please share the exact error messages or logs you ran into.
I'd love to be able to run this, but have a 3090 (ampere, sm86) - and only Hopper/Blackwell (sm90/sm100)kernels are shipped currently for the sglang-omni builds - is there any chance of an ampere build in the future?
I'd love to be able to run this, but have a 3090 (ampere, sm86) - and only Hopper/Blackwell (sm90/sm100)kernels are shipped currently for the sglang-omni builds - is there any chance of an ampere build in the future?
yes
I'd love to be able to run this, but have a 3090 (ampere, sm86) - and only Hopper/Blackwell (sm90/sm100)kernels are shipped currently for the sglang-omni builds - is there any chance of an ampere build in the future?
yes +1