search

erew123 / alltalk_tts

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.
GNU Affero General Public License v3.0
1.08k stars 115 forks source link

Deepspeed CMD or Activation (minor) Issue #20

Closed altoiddealer closed 10 months ago

altoiddealer commented 10 months ago

For starters - Alltalk and Deepspeed are working perfectly fine as intended.

I've added the following line to settings.yaml in order to auto-start Deepspeed, and this is also working:

alltalk_tts-deepspeed_activate: True

Now, onto what is not working as expected... please close this if you deem this outside the scope of your project, and if so I apologize for taking up your time.

I maintain a discord bot that integrates with textgen-webui, and it applies any settings defined in settings.yaml, or added by the bot script directly to module shared.settings (It imports the shared module, finds settings.yaml and adds all the settings).

The settings that I use are all working just the same as if I launch the webui directly.

If I change one of the shared.settings parameters, it is immediately reflected in the responses.

It's working great. All except for deepspeed_activate (specifically with this bot script).

For all I know, it may actually be applying Deepspeed but just logging "False".

Here is a screenshot of the cmd output in my script. I print shared.settings

screenshot

I'm not expecting you to work out and resolve any potential errors with my script - However, I've checked what I can and everything looks like it should work...

diagnostics.log

AllTalk was updated: 12/24/2023 Custom Python environment: Same env as textgen-webui Text-generation-webUI was updated: 12/24/2023

erew123 commented 10 months ago

Ah.. you don't need to put the -deepspeed_activate: True setting in for Alltalk to use DeepSpeed.

I'm sure I wrote about this somewhere in the notes... long story short, remove the --deepspeed setting from your settings.yaml.

You will set the AllTalk DeepSpeed in the settings page for AllTalk. That's the only place you need to set it.

image

As long as you have the correct DeepSpeed wheel installed, which for you is the Python Version: 3.11.5 and CUDA 12.1, all should be good.

But let me know if not.

erew123 commented 10 months ago

When I say thats the only place you need to set it, I mean if you want it to load up each time you start AllTalk. Otherwise, you can click the checkbox in the panel in the main interface.

image

Hope that sorts you out, but let me know if not.

altoiddealer commented 10 months ago

Oh, wow - I didn't notice all those prominent screenshots in the main page and that link in textgen-webui to access the main settings panel :)

I removed that param from settings.yaml, toggled the setting, and now it is indeed working in all places by default including my bot script.

Thanks! And amazing work on this, please keep up the good work :D

On a side note, the only voice I personally clipped myself for coqui / alltalk was David Attenborough, and it was actually from the same interview that the one in your voice package originated from. However, I think I clipped the best portion of the interview where he doesn't stutter. https://mega.nz/file/XIoWlSIS#IUZRUFgprIwAm6NVCewCHE_5WWdK18wB0tTLRxdY9bk