Spaces:

mrfakename

/

E2-F5-TTS

like 1.82k

Running on Zero

Hi,
It is not recommended to use this space as an API, as the rate limits on ZeroGPU are quite limited.
I know some other services such as Replicate or FAL are offering API usage as a paid service though.

Hi. What should I do in case I want to deploy this model using an API using a self hosted VPS (KVM) using docker?
Thanks in advance!

@jmoralespestana If you'd like to run it locally as an API, you could either run the Gradio space and use the Gradio API (probably not for production) or create a FastAPI server that wraps around the F5-TTS Python package. Note that in order to run it at a reasonable speed you will probably need a GPU.

Please tell me how to enable GPU for this?

@Haxxaan Are you running the model locally (on your computer) or in the cloud?

Yes I am running locally using Python coding, I took this model from your github

If you have CUDA + torch installed correctly it should work out of the box. Is it not utilizing the GPU properly?

It's not, for running this model I have enable GPU, but I don't know how to enable GPU

Do you have a GPU connected to your computer?

Yes, I have

Can you try running the following in the Python shell:

import torch
torch.cuda.is_available()

Hi @mrfakename : First of all thanks for this wonderful f5-tts library. I really liked the quality of the cloned voice and generated speech.
It will be great if you expose its inference API ( Fast API implementation) and host it in Litserve (Lighting.ai) as a community template.
Litserve is a scalable high-performance inference engine for AI models with a minimal interface and its completely ideal for dev to try, test & deploy at scale.