AllTalk is based on the Coqui TTS engine, similar to the Coqui_tts extension for Text generation webUI, however supports a variety of advanced features, such as a settings page, low VRAM support, DeepSpeed, narrator, model finetuning, custom models, wav file maintenance. It can also be used with 3rd Party software via JSON calls.
To answer the question of "Does AllTalk run on Fedora" and "Can you configure DeepSpeed" Ive run a test in a Virtual Machine.
Please note, that I did not setup the Nvidia CUDA Toolkit 11.8 so that I could perform Finetuning, I only used the Nvidia CUDA Toolkit 12.3 as it was simple to install and I cannot test Finetuning within a Virtual Machine. But Finetuning needs the Nvidia CUDA Toolkit v11.8 to be installed and correctly configured.
TLDR: Yes it works fine
All the steps below are from following:
QUICK SETUP - Standalone Installation
Linux DeepSpeed - Standalone Installation
from the main page, however, I did use Fedora's own NVIDIA CUDA Toolkit installation for simplicity.
Cloned the Github
Allowed atsetup.sh permission to run and ran it, selecting Standalone Installation and Option 1
Let the Install run through
Installation completed fine
Exited atsetup.sh and ran ./start_alltalk.sh to see if it would start up fine and it did
It was a little slow in a Virtual Machine, but it did start and load the model without errors
Next I went to compile and install DeepSpeed. To keep it simple, I followed the "Fedora 37 and later" instructions from here https://rpmfusion.org/Howto/CUDA
I ran the CUDA install from RPMFusion
Waited for that to complete
Followed through the Linux Standalone Instructions for DeepSpeed
Tested that it was working with nvcc --version
Ran the pip install deepspeed which successfully compiled
Started AllTalk with ./start_alltalk.sh. It was slow being on a Virtual Machine, however it detected AllTalk and loaded the model in correctly. It did give 1x error about NVML which is to do with the Virtual Machine environment
**Loaded up the Settings page and tested TTS, which worked fine bar another VM related error NNPACK***
To answer the question of "Does AllTalk run on Fedora" and "Can you configure DeepSpeed" Ive run a test in a Virtual Machine.
Please note, that I did not setup the Nvidia CUDA Toolkit 11.8 so that I could perform Finetuning, I only used the Nvidia CUDA Toolkit 12.3 as it was simple to install and I cannot test Finetuning within a Virtual Machine. But Finetuning needs the Nvidia CUDA Toolkit v11.8 to be installed and correctly configured.
TLDR: Yes it works fine
All the steps below are from following:
from the main page, however, I did use Fedora's own NVIDIA CUDA Toolkit installation for simplicity.
Cloned the Github
Allowed
atsetup.sh
permission to run and ran it, selecting Standalone Installation and Option 1Let the Install run through
Installation completed fine
Exited
atsetup.sh
and ran./start_alltalk.sh
to see if it would start up fine and it didIt was a little slow in a Virtual Machine, but it did start and load the model without errors
Next I went to compile and install DeepSpeed. To keep it simple, I followed the "Fedora 37 and later" instructions from here https://rpmfusion.org/Howto/CUDA
I ran the CUDA install from RPMFusion
Waited for that to complete
Followed through the Linux Standalone Instructions for DeepSpeed
Tested that it was working with
nvcc --version
Ran the
pip install deepspeed
which successfully compiledStarted AllTalk with
./start_alltalk.sh
. It was slow being on a Virtual Machine, however it detected AllTalk and loaded the model in correctly. It did give 1x error about NVML which is to do with the Virtual Machine environment**Loaded up the Settings page and tested TTS, which worked fine bar another VM related error NNPACK***