https://uberduck.ai/ logo
Join Discord
Powered by
# talknet-support
  • w

    WeegeeFan1

    12/15/2022, 11:29 PM
    Yea I'd seen that and thought that
  • w

    WeegeeFan1

    12/15/2022, 11:29 PM
    It may just be my training notebook is screwed up somehow
  • w

    WeegeeFan1

    12/15/2022, 11:29 PM
    I'll go back and retry with a new notebook
  • s

    Sonic2022_mario

    12/15/2022, 11:29 PM
    Can You Make A Meme Out Of My Results On Your Youtube?
  • w

    WeegeeFan1

    12/15/2022, 11:29 PM
    Results with what? The most recent files?
  • s

    Sonic2022_mario

    12/15/2022, 11:30 PM
    This
  • s

    Sonic2022_mario

    12/15/2022, 11:30 PM
    And This
  • w

    WeegeeFan1

    12/15/2022, 11:30 PM
    I'm not sure what to animate
  • s

    Sonic2022_mario

    12/15/2022, 11:30 PM
    Chuck E Cheese Monke
  • w

    WeegeeFan1

    12/15/2022, 11:30 PM
    If thats what your asking
  • w

    WeegeeFan1

    12/15/2022, 11:30 PM
    I'd rather not I have a lot of my own projects I'm working on
  • w

    WeegeeFan1

    12/15/2022, 11:30 PM
    Which are A LOT of work
  • w

    WeegeeFan1

    12/15/2022, 11:30 PM
    And I dont like taking requests because i have a dendancy of not being able to have the time to finish them
  • s

    Sonic2022_mario

    12/15/2022, 11:30 PM
    Monke Is 2nd Results And Narrator Is 1st Result
  • w

    WeegeeFan1

    12/15/2022, 11:31 PM
    I still cant do anything
  • x

    xomnow

    12/16/2022, 2:36 PM
    hello folks - I've got a question I'm hoping someone can assist with... I've trained four models previously back in the summer but now I'm having issues with step 6 - specifically the second portion of step 6 after durations.pt has been created where it seems to run each wav file... it's taking forever (I remember this step being a long none, but not this long) and I seem to be timing out in colab. I got like 80% of the way there and then colab dumped me (using colab pro) and when resuming this step the notebook knows durations.pt is done, but starts at the beginning of the files again... I've got about 6.7 hours of clips (2938 I believe) and is this just too much now? Will this step not complete because it appears it will take longer than 24 hours?
  • w

    WeegeeFan1

    12/16/2022, 9:19 PM
    So it sounds like even after running your step it cant finish with colab free gpu resources, and when you try to train it again, it runs out again, endlessly running out of gpu points and never training?
  • x

    xomnow

    12/16/2022, 9:29 PM
    I'm on ColabPro - and I've got 200+ compute units left, but I'm timing out (24 hours I guess) before the second portion of step 6 finishes - it's not a lack of compute, but that I can't get it completed in what appears to be a 24 hr limit - I don't recall this taking so long when I did this back in July (and I only mention the past because I've had success using this colab notebook, just not yesterday/today it seems)
  • a

    Andy Tyrer

    12/17/2022, 5:56 PM
    @xomnow are you using a A100? I am on Pro too but the colab notebook does not run and errors. What do you use?
  • x

    xomnow

    12/17/2022, 6:03 PM
    I'm getting T4s every time
  • w

    WeegeeFan1

    12/17/2022, 9:39 PM
    What would I do if I wanted to manually create the zip file for a talknet model?
  • w

    WeegeeFan1

    12/17/2022, 9:39 PM
    I used up all GPU credits in collab but can't save the model training into a usable file
  • w

    WeegeeFan1

    12/17/2022, 9:39 PM
    So I need to know what to copy and paste into a zip to create the zip file for the model
  • w

    WeegeeFan1

    12/18/2022, 4:05 PM
    How big is your dataset?
  • x

    xomnow

    12/18/2022, 4:20 PM
    It was nearly 7 hours (audiobook) - I pared it down to a bit under 5 hours and was able to complete step 6 and indeed finish the training. My question was more... has something changed with how step 6 proceeds? It seems interminably slow now. I remember the first portion of step 6 taking... I dunno, maybe like a couple of hours on large sets, but the second portion where it parses each wav (on it's own line) now takes ages where that used to be much much speedier.
  • w

    WeegeeFan1

    12/18/2022, 4:24 PM
    I don't think the notebook has even been updated since about May of this year, so no it hasn't changed.
  • w

    WeegeeFan1

    12/18/2022, 4:24 PM
    Nobody dares even touch it at all in fear of breaking everything
  • x

    xomnow

    12/18/2022, 4:29 PM
    I know the notebook is the same as when I used it in July, but Colab has also changed how it works since then, and it seems it just processes the files differently. One of my models from July was like 14 hours of audio and I had no issue getting through step 6. I guess - I was curious if others had noticed a change in step 6's behavior. I've tried figuring out how to get the notebook to run locally on my machine, and indeed, while I can't get it all to work (it can't find my gpus), it did blast through step 6 super fast. Like... two hours maybe whereas on Colab it timed out after 24 hours. I tried to cheat step 6 by running that part locally, uploading the files to Drive, then working from step 7 onward. That worked up until step 11 where I got errors - best I can determine is that running step 6 locally didn't quite work as far as step 11 on Colab was concerned. So, TL;DR - I pared down the dataset and got through it, but it sort of sucks that one can no longer use really large (and good quality) datasets for fear of timing out. I guess... it would be amazing if step 6, second portion, could save its progress so one could resume upon timeout.
  • w

    WeegeeFan1

    12/18/2022, 4:54 PM
    What it could be is that Colab has different limits nowadays whether you pay for colab pro or not
  • w

    WeegeeFan1

    12/18/2022, 4:54 PM
    Remember that AI tech has gotten MUCH MUCH more popular over the last year
1...495051...74Latest