RIP ELEVENLABS! Create BEST TTS AI Voices LOCALLY For FREE!

[@Aitrepreneur]

HELLO HUMANS! Thank you for watching & do NOT forget to LIKE and SUBSCRIBE For More Ai Updates. Thx

[@Refinement99]

How come you are not using ai speech in your videos? I'm not trying to be mean or judge or anything, but you have a strong accent and it's sometimes hard to understand. Even though I love your content

[@Sanguen666]

not bad, but tortoise is much more accurate imho. check out the video 'how charsi became a blacksmith' it was made with tortoise TTS.

[@trezero]

Would love to see how this can all be done programmatically as a next step.

[@germanher7528]

MEET LOCAL TTs NEAR YOU!!!! 1-800-HUGE-BOOBAS

[@zirufe]

Hi. If I would like to add another language for training the voice for TTS. Is there any workaround or other model to be used like in this video?

[@GGfrostZ]

Have fun in dependency hell trying to install this!

[@heady2905]

XTTS has such a good voice generation. If you repeat a sentence it sounds everytime different and if you combine it with your RVC voice model you got the best thing ever. The future of AI must be open source and it is good you show everybody how to use this powerful AI technology. Greetings from Germany 🙂

[@v11cu96]

Yeah I wouldnt say the 'sounding different each time', is a good thing when you want consistency though,Sorry for being negative, So far my open source TTS journey has not been great. I feel like im lucky if can generate 2 sentences in a consistent pitch and accent, maybe I need to try with an RVC model like you suggest?. Or just wait for the tech to impove a bit more.

[@MrRaja]

​@@v11cu96RVC method would be you recording the sentences how you'd like or using Xtts to generate a Audiofile to be used in RVC... My problem is how do I make it emotional like for example: What![Angry] Why didn't you tell me?[Desperate]

[@LJames-ez9lr]

@heady2905 oes this tts actually work for you? I got errors during installation and it wont launch.

[@MAST]

This comment sound like AI.

[@LJames-ez9lr]

@@MAST i figured it out! i only had Visual studio installed and not the tools

[@OctoberFox]

There's a lot of information he leaves out of his instructions in many of his videos (like missing requirements). Thanks to the other commenters for some of their input, as it's been really helpful.

[@Ull3Rnet]

Hey @Aitrepreneur - Quick heads up on the XTTS-RVC-UI install on Win 10. It installs protobuf 5.26.1 by default, which didn't work for me. Downgrading to protobuf 3.20.0 fixed the issue. Just thought this might help others running into the same problem!

[@yoniwoker]

I installed it, but the cmd window keeps closing. I press est, and cmd appears and closes quickly.

[@Ull3Rnet]

@@yoniwoker try editing the bat file, add a new line at the bottom saying Pause , save, then run it again. Then you should be able to see the error before the window closes :)

[@SuperFurias]

for everyone having setup py errors: run the install.bat file wait for everything to be installed see error close the cmd open a new cmd inside the folder type "call venv\Scripts\activate" type "pip install tts" wait for everything to be installed, close cmd then run again the install.bat file no more setup py errors. don't ask me why, because i don't know

[@StainCorb]

This finally solved my Finetune install, now I just need a couple of days to figure out RVC version, installing things through the command prompt is a party... ... lol

[@Rambo....]

👍Thanks man, you solved my problem here.

[@petepablogaming243]

What the hell. That fixed it. I also have no idea why though

[@diehgo_sp]

totally not working for me

[@SuperFurias]

@@diehgo_sp sounds strange, did you do it correctly? i know for sure that installing the tts package fixes setup py errors. but maybe you are having a different error, or simply did the procedure wrong, so could you tell me exactly what error are you having, and where?

[@vi6ddarkking]

Between this in six months, The SD3 Fintunes, The tools that finally are getting us to consistent characters, and the Lama 3 Fintunes. This year's Sillytavern video Is going to be bonkers.

[@user-jk9zr3sc5h]

SD3 sadly won't work with controlnet due to its lack of UNET architecture, but hopefully something similar is shipped soon

[@GengoSenmon]

How do we integrate this with SillyTavern so we can speak back and forth with the voice we generated?

[@wakegary]

i dont know what ure talking about but i cloned this depos and so um now I need to know what ure talking about. cuz this is rad and does tortoise level at bark speeds. god we're nerds. (sonic checking watch gif) also expecting open source udio clone (odel-may eak-lay) to get git got soon. hoping! is musicgen still the top dog?

[@ItsBrody]

Got hella complicated by the half way point. I wish you used a different persons voice, Obama's voice almost already sounds like a robot and the last sample you showed us honestly didnt sound that great.

[@derekthemagician]

Whenever I see his logo image I know I want to do what he's doing but I'm not going to be able to.. Lol

[@sir_no_name1478]

Lol for real? I mean you can stop the video when it is complicated and follow it step by step. And if it is really unclear just google the words he says. I would not use audacity to append the voice to a 2 minute clip though. There is a reason they want 2 minutes.

[@jvdome]

Yea, i think this still too hard to do, i could follow step by step but dealing with errors is not worth the hassle if i am doing this for fun

[@zachary3603]

@@derekthemagician This is the only thing that makes it worth doing rn. As soon as it becomes easy, it won't stand out much.

[@PACOBRYAN-cj9gf]

just summarize the vid with AI

[@Neyokah94]

This video is 17 minutes of "But wait, there's more!" and it's SOO good. Thanks!

[@carlosgonbr]

Unfortunately, it cannot be used, many errors occur during installation following the steps given. Code errors, things not found, different versions. things that are difficult for non-programmers to understand. what a shame.

[@futaa34]

you are correct buddy. only a handful could have it running

[@zytoh]

i started debugging with chatgpt, followed every step and error, and after 7 hours got it working, just dont give up, especially if its for a buisness which is why im using iut

[@carlosgonbr]

I discovered the main problem with my installation and after solving it, I installed everything without any errors. It's the Microsoft Visual C++ 14 package. It's not enough to just install Visual Studio, you have to install the package along with it, but it's not that intuitive. Look for a video "Fix: Microsoft Visual C++ 14.0 or greater is required in Python" from the "Hey, Let's Learn Something" channel that helped me. It's very simple. Then come back here and thank our friend from the channel who introduced us to this wonder.The three programs installed and are working without errors!

[@popipopi3126]

took me hours to fix everything even knowing what im doing

[@ChasingStars7111]

@@popipopi3126 how did you fix it ?

[@olucassantos]

I couldn't get past 2 minutes of tutorial, there were so many errors, some I solved, others were impossible to solve. 4 hours trying. Still, thank you very much for your efforts, unfortunately I gave up trying

[@intelligenceservices]

you're right, this is really badly maintained.

[@siddhantshahi5027]

took me 4 days to solve all the errors

[@sz9515]

@@siddhantshahi5027 do you have a simple steps process?

[@PD-THANH]

This is a game changer! Training my own TTS model locally seemed impossible before, but this makes it surprisingly achievable. Has anyone tried using this with longer audio samples, like an audiobook narration? Curious to see how the quality scales.

[@intelligenceservices]

did you get it to install and run?

[@the-papaw]

I can't get the second one to work (xtts-finetune-webui). I keep getting "error Connection errored out." when I try to do the "Step 1 Create dataset" Shows this in the DOS prompt "ERROR: Exception in ASGI application Traceback (most recent call last):

[@audiogus2651]

Woah, perfect timing on the vid, was just looking this stuff up today. Thanks homie!

[@nodewizard]

Goodbye Eleven Labs. They were overpriced and closed source. This XTTS model is amazing. Merci Monsieur Aitrepreneur. As a little goodbye kiss to Eleven Labs, I'm going to clone my favourite voices that they have. Lol.

[@Aitrepreneur]

Have fun ;)

[@AGI-Bingo]

Aaaaand 11Labs dropped SOTA SongGen

[@GengoSenmon]

How do we integrate this with SillyTavern so we can speak back and forth with the voice we generated?

[@wakegary]

Sucks that they didn't put One Lab aside for something more traditional like textile manufacturing.

[@zachary3603]

Which python version did you use for this? Trying to get deepspeed to work, but it's saying Python 3.9 might be too high of a version xD

[@coloryvr]

Cool! I was waiting for this! Happy colored Greetinx!

[@geneanthony3421]

Been loving your videos. AI is moving so fast anymore and I like to keep up.

[@Mikes-Code]

Requested float16 compute type, but the target device or backend do not support efficient float16 computation.

[@andro1234567890100]

1. Open the xtts_demo python file using an IDE 2. change: if torch.cuda.is_available(): compute_type = "float16" else: compute_type = "float32" to: if torch.cuda.is_available(): compute_type = "int8" else: compute_type = "float32" 3. save Took me a whole day to figure out. Just needed some sleep and then went and read through the git repository's reported issues and found it in there. Seems to be working for me now.

[@spejarn]

Having a 8GB card running LLM, SD and this thing for some roleplaying in Silly Tavern is getting silly. I hope next gen low-mid range cards has a decent amount of VRAM.

[@ItsOk-mq9ex]

4060 has 16gb variant, 5060 will probably be the same just higher cost.

[@snark567]

Got a refurbished 3090 just for the ram.

[@TomiTom1234]

Awesome.. I always appreciate TTS and Voice cloning videos

[@LJames-ez9lr]

@TomiTom1234 did this actually work for you? none of them work for me after installing.

[@TomiTom1234]

@@LJames-ez9lr Two did work, but XTTS-RVC-UI didn't, I get a red error after installing it, sadly.

[@zachary3603]

@@LJames-ez9lr What errors are you getting?

[@intelligenceservices]

@@zachary3603 try installing it yourself?

[@Van_of_the_lake]

You are the AI mentor of this generation, thanks for the hard work

[@vi6ddarkking]

Say Our AI overlord. A mad lad that goes by gradientai on huggingface. Made Llama 3 work with a 1 million+ token context size. Since you have a 4090 now. I am curious if you could see how far it can go on our local machines.

[@DarthyMaulocus]

tried this got it working and i know how to get it to work on pretty much any machine now. must say you provide useful information however you miss out so many details that are necessary(you make no mention of the python versions or cuda for example but again anyone actually interested in this will persevere for a while), I managed to get it up and running partially due to your help much appreciated anyways

[@magenta6]

Kudos to Coqui TTS for making this available!

[@DiffusionStudio4k]

They are out of business unfortunately 😢

[@errorgradov8050]

damn,fine-tuning was really heavy problem for me,now i got it thanks

[@redmatwinch]

12:45 where didi you take the index and pth file?

[@metanulski]

I like the Videos but the instalations never work. :-(

[@JRis44]

What system you using and where you having issues?

[@metanulski]

@@JRis44 windows, but I dont remember the exact error.

[@JRis44]

@@metanulski have you attempted to try any of the ai apps lately? if you have brave browser we could do a walk through. I have some time today and dont feel like doing anything overly productive and dont mind helping someone today or this weekend maybe.

[@varun3771]

@@JRis44hello if you could help me it would be good. I have alr torch 3 or higher installed but teh program insists on having torch 2.1.1

[@swannschilling474]

Thanks so much for this one! Cannot wait to try!! 😊

[@noobicorn_gamer]

For once, I’m not clickbaited and I’m happy i dropped by

[@lucifer9814]

After watching tons of such videos about these AI tools, I realized the ones making these tutorials, especially if they're programmers assume the whole world to be programmers as well. " Install this, install that, install this fucking shit ", he says, half of them don't even end up bloody installing. ALL I REALIZED IS THAT CODERS AND PROGRAMMERS HAVE NO PATIENCE WHATSOEVER AND DON'T TRY TO UNDERSTAND OTHERS FROM A LAYMAN TERM.

[@Darfail]

eh believe me programmers struggle too to install anything, it's a perpetual hell

[@lucifer9814]

@@Darfail LoL

[@juleslincredule]

Great stuff! Anything for Mac users? Just asking... 😃

[@latemanparodius5133]

As I'm chatting in SillyTavern, I notice that the command windows try to reference emotion voice models, such as joy.pth with joy.index or surprise.pth with surprise.index. Sure, it still works without them, but do you know if those will have to be custom trained models for that character in that emotion, or is there some generalized emotion model somewhere that can be copy/pasted to multiple characters?

[@anoirbentanfous]

Now, develop this into an API for generating real-time text suitable for browser reading and audiobook listening. Ensure multilingual support, accurate number pronunciation, and handle various cases like omitting annotations and URLs.

[@tanjabeckers9478]

Merci ! Genial pour cette version Open ❤source

[@creed4788]

Xtts fine tuning not working for all people! This tutorial is breake

[@MrPer4illo]

Great job 👍 How about customizing LLM next?

[@Lil_Shoosh]

3:23 1st Method 5:44 2nd Method 10:15 3rd method

[@planetmuskvlog3047]

Yeah, but can it do foreign languages as easily as Eleven labs ‘multilingual v2 or v3?

[@FluorescentApe]

Is there a V3 for elevenlabs? I only se V2.

[@planetmuskvlog3047]

@@FluorescentApe there’s “multilingual v3” now

[@FluorescentApe]

@@planetmuskvlog3047 what's weird. Can't see it. Maybe only a select portion of people can use it?

[@tapikoBlends]

amazing!

[@jurandfantom]

Damn, I was hoping to hear/see something better in terms of quality since ByCloud/Jared videos. I hate to see such stagnation :/ Thanks AiT for update on the topic

[@DanielPartzsch]

Could you increase the tts quality and likeness of the second model even more with a longer audio clip than 2 minutes? Or doesn't it make any difference above this length?

[@Jorvanius]

I'm wondering the same thing. Did you test it? 👀

[@MrRaja]

I think someone installed a A.I. brain chip without me knowing like while i was asleep... For some reason i understood every single word in this video...😂😂

[@Alice_Fumo]

Hell yes, this is exactly what I wanted!

[@mauricioermel]

Why erverytime when I install XTTS FineTune it does not create the two folders base_models and finetune_models? When I run the start.bat it opens, but obviously I am not able to train any model.

[@yngeneer]

sooo.....can it be stitched to the silytavern somehow?

[@42ndMoose]

sillytavern already has a way of adding xtts extention, which has live realtime streaming. you can find that in the plugins tab in sillytavern. but then you'd have to go through a complicated process, at least for me. to put ST in staging etc

[@duck-tube6786]

I wish K covered off exactly this. How do you take your Uber TTS model and then run it in Sillytavern

[@yngeneer]

@@42ndMoose is there a tutorial for that?

[@LeCamionDAmar]

Unfortunately the tutorial is outdated and nothing works anymore. sad :( Ok : Still works but i have to install dependency manually dont know why

[@Tokaint]

Do these have python integration? Like instead of using and downloading the webui could I jusr use code to tell it what to generate and save the file somewhere? And if yes where do I go to find out how to do that?

[@obamagaming7909]

Would it be possible to integrate this into a python script?

[@davidsmith-lv4kq]

how much vram needed?

[@rachkaification]

8:40 The reference audio sounds way better and closest to Obama's voice than the generated audio from it.

[@kritischinteressiert]

Thank you for the great video ! Is there a proper ATS as well? Or an app, that does Dubbing like 11Labs?

[@redt1903]

Nah bro ima use this for brainrot edits🔥🔥🔥🔥🔥

[@Ryzza5]

Any good tutorial video needs links to the required downloads.

[@futaa34]

exactly

[@loszhor]

Thank you for the information.

[@Kujamon]

I get "ERROR: No matching distribution found for torch==2.1.1+cu118" when running the install.bat, despite intalling the pre-requisities

[@LuminRL]

ever find a fix??

[@Kujamon]

@@LuminRL Nope

[@DarthyMaulocus]

you need python 3.10 and set up path. ive got it all working any questions ask me. Its also in the issues page of github number 8 I also made a thread thats completed.

[@LuminRL]

@@DarthyMaulocus goat. once I get home I'll try again and see if I can work it out

[@andro1234567890100]

@@DarthyMaulocus This worked. Thank you! For anyone going through the same struggle as me, don't try to download the zip files for < Py3.10.11. Py3.10.11 was the last release with a download link. Find that and you'll find the installer link.

[@thomasroyer5017]

is there a difference between your github and the original xtts-webui ?

[@nicotvupa]

Thanks! I found the colab for the first one. Are there colabs for the other 2?

[@ash3844]

colab link pls

[@adriancoleman2876]

I cant wait till AI can recognize heavy reverb. i have ripped all of Dr Brackmans voice files from Supreme Commander in anticipation for just that day.

[@zealgaming8161]

I've been waiting to resurrect the late Tony Jay's work on The Transcendent One from Planescape Torment since forever. Highly recommend you check him out if you want a really scratchy, dark, evil god voice.

[@ArabianShark]

Awesome video! I had been waiting for just this for ages! Thank you very much!

[@thanesbusiness5001]

after two hours, i still can't get it to launch. it opens then closes

[@mariokotlar303]

Is this approach scalable to large text sizes? Like, if I tried to TTS an entire book, would that take infinite VRAM or endless dealing with 2 minute chunks or something, or would it just work?

[@DaveGamesVT]

I'm guessing this doesn't work on AMD cards?

[@nixaristix1819]

Thanks! Can I use these methods for audiobooks with synthetic voices?

[@-Burs]

Cool stuff, I just wish more languages are supported.

[@ssw4m]

It couldn't be too difficult to find two minutes of Obama speaking. Why not spend a few minutes getting a longer sample, and presumably get even better results? Thanks for the demo, anyway, it's awesome tech.

[@arete_]

15:48 for end result

[@JohnRiley-r7j]

Wow a finetuning WEBui is awesome,in all other apps my 8gb Vram GPU was not nearly enough for training but with this Vram usage is like you said minimum,training is pretty fast and quality is amazing! One question,what if you want to train one model to be good with multiple voices,is that even possible or you need to train new model with every new voice you are using? Thanks!

[@GraveUypo]

i had a setup with tortoise tts + rvc, but this seems better. thankfully it also works on linux, form just watching the video i thought it might not. my tortoise thing doesn't. i'll try it later.

[@kylegeib9161]

Wasn't a very good idea to repeat 37 seconds of reference audio at the start. With all the time I've used 11Labs' solution, even the ultimate version you have here doesn't sound as good as even their English V1 model.

[@Aitrepreneur]

yeah it wasn't a good idea, just me being lazy but it still worked ok. Not sure I agree with the final result, it's very similar to an elevenlabs quality and it's free and unlimited, if you want to pay to use 11labs it's your choice, I'm giving another possibility to people who can't afford it or just want to save money for a very similar level of quality

[@InvadeNormandy]

Mines not working and keeps spitting out gradio errors despite following the instructions to the letter. webui and finetune both.

[@andrerd99]

Great video man! is there anyway possible that i can tts and transform the audio directly to my mic on discord for example with this?

[@claytaan]

Do i need to install CUDA on my pc as well?

[@digitalface9055]

now I would really like to see tutorial how to fine tune your own language model and utilize it in LLMs.

[@DanielPartzsch]

Why do you not just use the rvc enhancement option in the xtts WebUI directly? Is it slower or of lesser quality compared to the full RVC version?

[@thays182]

Are the outputs from these methods something that can be used for speech to speech in something like W-Okada? Or is that a different process?

[@rickyparker2943]

What do I do if I get this error message? ERROR: Could not find a version that satisfies the requirement torch==2.1.0 (from versions: 2.2.0+cu118, 2.2.1+cu118, 2.2.2+cu118, 2.3.0+cu118) ERROR: No matching distribution found for torch==2.1.0

[@annonymat]

Same here!

[@kevins_campfire]

I get this error as well. google isn't being helpful so far

[@vaughanbury1]

@@kevins_campfire change 2.1.0 to 2.2.0 seems to be installing then

[@AltMarc]

@@kevins_campfire it probably has to do with the cu118, do you have Cuda version 11.8 installed? you can try to cheat it, by modifying the requirement file...

[@FluorescentApe]

@@AltMarc I have Cuda 11.8 installed and added to my path, but still doesn't work :/

[@j0hngk139]

Hi, do you plan to make a video on how to use LLMs and other AI things using ROCm for Radeon and Windows users?

[@Aitrepreneur]

well I don't have any AMD GPU so can't really show that

[@ts757arse]

I've settled on the fact that I either have to wait a few months for AMD (or quants to enable CPU) support or get an Nvidia card. Nvidia is still on my shit list and I won't be doing that. The company's practices haven't changed.

[@ArmoredAnubis]

Does this work in Silly tavern?

[@GengoSenmon]

This is what I wanted to know the entire video. How do we integrate this with SillyTavern so we can speak back and forth with the voice we generated?

[@grayhamgrayhamson1466]

Perfect timing!

[@iamnobody-001]

1:54 , Hi, you said to install the c++ visual studio but after i download and try to install, it offers so many option, which one do i have to install? or just the visual studio core editor only ? thanks for your help

[@katze_ksb]

fine tuning xtts encoder unfortunately throws errors ( Library cublas64_12.dll is not found or cannot be loaded)

[@syedrahman1714]

same happening

[@Mateo906]

When trying to install torch i get this error: ERROR: Could not find a version that satisfies the requirement torch==2.1.0 (from versions: none) ERROR: No matching distribution found for torch==2.1.0 Does someone knows why this happens?

[@sneett7670]

Thats because the python version is too new. just download the 3.10 one. I found this out the hard way.

[@NT-eh6om]

I was able to do it by installing torch and torchaudio seperately.

[@intelligenceservices]

​@@NT-eh6om where did you install those? globally, to an anaconda venv, or to the xtts venv? the installer itself is a mess which takes the confidence down a notch.

[@regeneric59]

Does this work with any gpu? Or just nvidia

[@ArthurHuizar]

I have AMD and CUDA is not having it.

[@falco2911]

3:27 no module named 'requests'

[@rhythmradius]

I get lost at the 6:00 point when you start talking about xtts-finetune-webui. Where is that supposed to be?

[@rhythmradius]

Never mind. I got it! :)

[@anagnorisis2024]

Can this generate emotional voices like angry, sad, happy? Or it's basically just different voices but the intonations are more or less fixed?

[@alexanderg8466]

I had an error in the last step of downloading: "ImportError: DLL load failed while importing transformer_inference_op: The specified module could not be found."

[@zSiuu]

Were you able to solve it?

[@streamy73]

Same issue

[@SmexMyPocky]

I don't understand the point of the last step? Why are we going through RVC when it doesn make a model, just to download a reference wav?

[@jello195]

@Aitrepreneur This makes me wonder, if it is possible to for length and if there are shortcuts to auto translate voice to voice (in the optics of easy dubbing). I'd be interested in a video about that if ever such app exists.

[@blackpantherAI]

where i can find AI voices already made by the community?

[@Aitrepreneur]

google rvc models

[@dumbsurvivor1]

@@Aitrepreneur why don't you give the link in the description it's like you're trying to finess a little bit

[@Qubot]

Thanks for the tutorial, however here is 3 env, is it possible to put them all in the same env ?

[@geneanthony3421]

Not sure why, but finetune seems to silently drop when trying to create a dataset. It runs until 100% and the process just closes without an error. Anyone else run into this issue?

[@DestinyFaux]

Couldn't even get it to run lol

[@geneanthony3421]

@@DestinyFaux found out that my issue was that I was running in Conda not venv. Seems to work that way

[@wedding_photography]

Definitely not as good as ElevenLabs, but not a bad result. Wish you had more examples, different speakers.

[@sohaibmoussaidelidrissi243]

Can you do a method for live voice change ? From voice to voice, thank you for the great video.

[@dziku2222]

Doesn't work. At 2:21 call install.bat I get this message: Nie mo'venv' is not recognized as an internal or external command, operable program or batch file. 'pip' is not recognized as an internal or external command, operable program or batch file. 'pip' is not recognized as an internal or external command, operable program or batch file. Install deepspeed for windows for python 3.10.x and CUDA 11.8 Nie moInstall complete. Press any key to continue . . .

[@GragSpcX_AI]

Same here! I don’t know how to fix this issue. I’m a beginner!

[@PresidentofAntifa]

You in the correct folder?

[@studiomusicflow4644]

so. it needs NVidia to run? any way to run on cpu or amd?

[@komakaze1]

I'd like TTS voices to act. Imagine giving it a long story and it would whisper, shout, laugh, cry, express surprise, embarrassment, bravado, fear, courage, disgust, curiosity and interest through voice. Can any TTS AI do this yet?

[@DarthyMaulocus]

yes im working precisely on this, it requires using different models from memory, i mean sure you can switch out models hence have emotions, its just delay of switching which is the worry

[@Darfail]

new advanced voice mode on chatgpt does that and...it's amazing

[@landyandy]

Hi alien. Do you have any tips for making XTTS WebUI voice2voice work on an older 1070 GPU? I load the voice model, choose the language and then generate. Error --> ValueError: Requested float16 compute type, but the target device or backend do not support efficient float16 computation. Bye and ty

[@andro1234567890100]

1. Open the xtts_demo python file using an IDE 2. change: if torch.cuda.is_available(): compute_type = "float16" else: compute_type = "float32" to: if torch.cuda.is_available(): compute_type = "int8" else: compute_type = "float32" 3. save Took me a whole day to figure out. Just needed some sleep and then went and read through the git repository's reported issues and found it in there. Seems to be working for me now.

[@ROUNAK275]

Can it work without gpu

[@benedictsforester7045]

finetune spits out errors like crazy. wasn't able to finish a single training