emikulic

6 ms·

1. 1.▲
   by emikulic 3y ago

   It's over for decels.
2. 2.▲
   by emikulic 3y ago

   Three epochs means it sees each token three times. The dataset is ~1T like you said.
3. 3.▲
   by emikulic 3y ago

   I want flying cars and moving sidewalks like I was promised by The Jetsons. :(
4. 4.▲
   by emikulic 3y ago

   Try https://huggingface.co/ehartford/WizardLM-7B-Uncensored and related models. They're not even trained on smut, just the neutered responses were removed before the RLHF stage (IIUC)
5. 5.▲
   by emikulic 3y ago

   I used to be bad at math but then I did a 360° on that.
6. 6.▲
   by emikulic 3y ago

   Emad said SD cost $600,000 to train. I wonder if Midjourney also had to pay that to train from scratch.
7. 7.▲
   by emikulic 3y ago

   What benefits does the Huggingface diffusers(?) implementation have over A1111?
8. 8.▲
   by emikulic 3y ago

   I like to run A1111 in --api mode and write my own script to drive it over HTTP.
9. 9.▲
   by emikulic 3y ago

   Wow, Reddit is going to lose a lot of pagerank.
10. 10.▲
    by emikulic 3y ago

    You're thinking of archive.is
11. 11.▲
    by emikulic 3y ago

    You mean the one that defeated two superpowers?
12. 12.▲
    by emikulic 3y ago

    Last time I looked into this, the answer was "because huggingface transformers and torch.load are written to do it this way" You could absolutely do something streaming, or mmap the weights instead of loading them into system RAM.
13. 13.▲
    by emikulic 3y ago

    This used to be the case but nowadays you can just pip install e.g. https://libraries.io/pypi/nvidia-cuda-runtime-cu11
14. 14.▲
    by emikulic 4y ago

    I hope you will run your own locally instead of letting Midjourney block and/or mess with your prompts.
15. 15.▲
    by emikulic 4y ago

    If edge devices includes my gamer PC then yes. For apps I'd recommend https://github.com/AUTOMATIC1111/stable-diffusion-webui
16. 16.▲
    by emikulic 4y ago

    If that's what paywall means now then sorry for the noise. :)
17. 17.▲
    by emikulic 4y ago

    I thought Emad said it cost $600k in GPU time.
18. 18.▲
    by emikulic 4y ago

    You have to create an account and agree to some stuff and be logged in but you're correct that you don't have to pay.
19. 19.▲
    by emikulic 4y ago

    The absolute state of the free and open web in 2022.
20. 20.▲
    by emikulic 4y ago

    Same. I've found the biggest problem was SNAT rewriting the (source) port number. netfilter, by default, doesn't do this. pf does but you can configure it not to.
21. 21.▲
    by emikulic 4y ago

    You can generate things the author doesn't like. And since you're doing it on your video card at home, nobody can stop you.
22. 22.▲
    by emikulic 4y ago

    Sorry to hear it, man. :( I hope you get better.
23. 23.▲
    by emikulic 4y ago

    Craiyon certainly doesn't.
24. 24.▲
    by emikulic 4y ago

    You can run craiyon / dalle-mini on a card with 8GB of VRAM if you decrease batch size to 1 and skip the CLIP step. Takes about 7 sec to generate an image on a 3070. I started with https://github.com/borisdayma/dal
25. 25.▲
    by emikulic 4y ago

    How long a network outage are you allowed to have before you lose your stake?
26. 26.▲
    by emikulic 4y ago

    IIUC if the CPU changes even once, your program jumps to (is context switched into) a recovery sequence.
27. 27.▲
    by emikulic 5y ago

    Or CNBC.
28. 28.▲
    by emikulic 5y ago

    Great episode. I rewatched it recently and it was just a bit too real.
29. 29.▲
    by emikulic 5y ago

    I wish Australia had spent a fraction of that money building quarantine facilities instead. :)
30. 30.▲
    by emikulic 5y ago

    Vouch takes the comment out of dead status. I then see other dead comments by the same user so I don't think it takes the user out of dead status.