Config page provides no way to load a fine-tuned model

[Artem-B]

https://github.com/erew123/alltalk_tts/blob/fff01107c89cea9e7f0405e247da0f39637dc957/system/at_admin/at_settings.html#L249

Similarly there's no way to select finetuned model via SillyTavern extension.

If there's an existing way to load finetuned model and make it available in SillyTavern, I didn't manage to find it. Updating documentation would be helpful.

[erew123]

Hi @Artem-B

I've had a quick look and it must be something that has changed. I know the way variables are saved changed in ST so maybe its related. Certainly AT is sending over to ST that a Finetuned model is available:

Its just not opening the option in ST now. As I say Ill have to take a deeper dive on this one and try see what's occurring with that.

As a temporary workaround for you, as long as you have the Finetuned model stored it the trainedmodel folder, it should be detected on AllTalk startup:

In a command prompt/terminal, you can manually force a load of the model with:

curl -X POST "http://127.0.0.1:7851/api/reload?tts_method=XTTSv2%20FT"

That would at least get you 80% of the way there.

As I say, Ill have to take a look at what ST is or isnt doing that stopping it presenting the option on the dropdown list.

Thanks

[Artem-B]

Finetuned model is detected, but it does not get loaded via ST UI.

[AllTalk Startup] AllTalk Github updated : 6th April 2024 at 10:45
[AllTalk Startup] Finetuned model        : Detected
[AllTalk Startup] TTS Subprocess         : Starting up
[AllTalk Startup]
[AllTalk Startup] AllTalk Settings & Documentation: http://127.0.0.1:7851
[AllTalk Startup]
[AllTalk Model] XTTSv2 Local Loading xttsv2_2.0.2 into cuda
[AllTalk Model] Coqui Public Model License
[AllTalk Model] https://coqui.ai/cpml.txt
[AllTalk Model] Model Loaded in 7.87 seconds.
[AllTalk Model] Ready

Manually loading finetuned model with curl works. Thank you for the hint.

AFAICT, the ST extension never sets tts_method=XTTSv2%20FT All I can see in chrome console is http://localhost:7851/api/reload?tts_method=XTTSv2%20Local.

[erew123]

Hi @Artem-B

I managed to figure what it was and have fixed it:

If you go here: https://github.com/erew123/alltalk_tts/blob/main/system/st_files/alltalk.js and click the download button:

And you will need to save that file over the top of alltalk.js in \SillyTavern\public\scripts\extensions\tts

All should be working! Ill have to send a PR over to ST to get it updated there at some point.

Thanks

[Artem-B]

https://github.com/erew123/alltalk_tts/blob/fff01107c89cea9e7f0405e247da0f39637dc957/system/st_files/alltalk.js#L284

Looks like we're comparing with a string 'true' instead of a boolean true here.

Update: looks like it's exactly what you've fixed. :-)

[Artem-B]

I can confirm that XTTSv2 FT is now visible in ST, and that finetuned model gets loaded when it's selected.