Doesn't work on CPU locally.

#1
by Rhueue - opened

When I try to clone this space and run it on kali Linux with all installed dependencies it loads the model then gives the Attempting to deserialize on Cuda error. I'm using the exact app.py in this space, yourtts runs fine on the same kali Linux Vm.

Oh, can you send me your console output, machine info (like cpu, ram) and kali version ? I’m new in AI, I may do somes mistakes, probably you will recv my futur response 3 / 4 day after 25/09/23. (college travel)… I will do somes research however, I need infos that I ask before.

Oh, can you send me your console output, machine info (like cpu, ram) and kali version ? I’m new in AI, I may do somes mistakes, probably you will recv my futur response 3 / 4 day after 25/09/23. (college travel)… I will do somes research however, I need infos that I ask before.

I use kali Linux 2023 in virtualbox with 14gb ram on an i7 7th Gen laptop with 16gb ram and gtx 1050 ti 4gb but kali can't use the gpu. When I clone your space and execute python3 app.py the model gets loaded, I reach Using model:XTTS then after 1 minute I get the Attempting to deserialize on Cuda error.

As I mentioned before Your_TTS runs very well on this same machine. I'm busy with college too and I understand you👍

Ok, I may have a theory, I've just had a bit of a think and there are a few questions to ask about the installation,

  • So you're on Linux, with just one CPU.
  • The yourTTS model works perfectly on this same machine, but the quality is poor.
    The question to ask is this:
    When did you install the TTS github repos?
    The fact that the xtts_v1 model is cpu compatible is very recent (3-4 days), the developers have modified the github repos on the 'dev' branch, which contained the cpu compatible version, if you downloaded it more than a week ago, your dependencies are not up to date, if you want to be sure you're on the right one you need to install the 'dev' branch and not 'main'!
    Try the following command on your machine:
    pip uninstall TTS
    or
    pip uninstall -y TTS'.
    Don't hesitate to add 'python -m pip uninstall TTS' if you have several versions of python.
    Then, to be sure, reinstall the requirements.txt file in my repository, i.e. :
    TTS@ git+https://github.com/coqui-ai/TTS.git@dev'.
    or clone the github ('dev' branch) and install it via setup.py if that doesn't work.

Ok, I may have a theory, I've just had a bit of a think and there are a few questions to ask about the installation,

  • So you're on Linux, with just one CPU.
  • The yourTTS model works perfectly on this same machine, but the quality is poor.
    The question to ask is this:
    When did you install the TTS github repos?
    The fact that the xtts_v1 model is cpu compatible is very recent (3-4 days), the developers have modified the github repos on the 'dev' branch, which contained the cpu compatible version, if you downloaded it more than a week ago, your dependencies are not up to date, if you want to be sure you're on the right one you need to install the 'dev' branch and not 'main'!
    Try the following command on your machine:
    pip uninstall TTS
    or
    pip uninstall -y TTS'.
    Don't hesitate to add 'python -m pip uninstall TTS' if you have several versions of python.
    Then, to be sure, reinstall the requirements.txt file in my repository, i.e. :
    TTS@ git+https://github.com/coqui-ai/TTS.git@dev'.
    or clone the github ('dev' branch) and install it via setup.py if that doesn't work.

Yourtts has an acceptable quality actually, plus I cloned your space today and installed the requirements and I updated tts to the latest version which is 0.17.4

So, It’s didn’t work again ?
I’m verry confused too, In my previous test on windows 10, 11 and linux (only on hugging face) I don’t have this problems, I’m just starting in AI, I advice you to go and make a ticket on TTS github, (I don’t have the anwser, sorry again).

Olivier-Truong changed discussion status to closed
Olivier-Truong changed discussion status to open

So, It’s didn’t work again ?
I’m verry confused too, In my previous test on windows 10, 11 and linux (only on hugging face) I don’t have this problems, I’m just starting in AI, I advice you to go and make a ticket on TTS github, (I don’t have the anwser, sorry again).

There are 3 issues about this on github and they closed them yesterday with no clear fixes to the issue. I'll uninstall TTS and have a fresh install of it starting from the dev repo again I'll see if it works this time but I highly doubt it. Thanks 🌹

Oh, Ok, so I think an install on windows (10 / 11), it’s the latest solution…
For me It’s work perfectly, one think you must install ‘ffmpy’ by it’s github repos with ‘setup.py’ and another (same problem at from import version. If you have this type of error, install this with the github repos and ‘setup.py’.
If you don’t have admin privilleges use the portable version of python and release < python3.10.6, else you have somes errors, for numpy change the python version < 3.10 to <= 3.10 in TTS requirements.txt, I understand that is a little archaic, but If you are familliar with python, you will found, I don’t have another solution, else use directly in hugging face. Thank for your understanding.

Oh, Ok, so I think an install on windows (10 / 11), it’s the latest solution…
For me It’s work perfectly, one think you must install ‘ffmpy’ by it’s github repos with ‘setup.py’ and another (same problem at from import version. If you have this type of error, install this with the github repos and ‘setup.py’.
If you don’t have admin privilleges use the portable version of python and release < python3.10.6, else you have somes errors, for numpy change the python version < 3.10 to <= 3.10 in TTS requirements.txt, I understand that is a little archaic, but If you are familliar with python, you will found, I don’t have another solution, else use directly in hugging face. Thank for your understanding.

👍🌹

Sign up or log in to comment