NanoGPT: The simplest, fastest repository for training medium-sized GPTs • Hacker News Board

Re: NanoGPT: The simplest, fastest repository for training medium-sized GPTs

#21

Post by mromanuk » Wed, Jun 12, 2024, 12:20 AM UTC

VagabundoP wrote: post #7

Would any GPU - even an old one - work for training these models? I have a bunch of little home apps/ideas for some specific training I'd like to do, but don't currently have any recent GPU of note. Just a few old ones in my attic somewhere. EDIT: ah someone has posted a blog post in the comments with some info EDIT2: to be clearer I was talking about finetuning. I assume thats quicker/less intensive. I really need t…

Yes, fine tuning is fast, cheap and "easy", you can do it in a no so expensive GPU without issues.

Re: NanoGPT: The simplest, fastest repository for training medium-sized GPTs

#22

Post by serverlord » Wed, Jun 12, 2024, 4:13 PM UTC

paradite wrote: post #18

serverlord wrote: post #16

We need a simpler way to make GPT get mass adoption and I don't think GPT Store from OpenAI will do that. We need a unique interface with a standardized process to help people make use-case-specific GPTs. Let's see what the future holds.

I actually built a purpose-made UI (desktop app) for coding using ChatGPT, because I found that chat interface is not ideal for daily coding tasks. Curious what you think about it: https://prompt.16x.engineer/

This looks super interesting. Thanks for sharing! I will share my thoughts in a couple of days.

Re: NanoGPT: The simplest, fastest repository for training medium-sized GPTs

#23

Post by alok-g » Wed, Jun 12, 2024, 5:15 PM UTC

paradite wrote: post #17

alok-g wrote: post #15

Earlier quoted context omitted.

Thanks a lot for sharing this. How much does the net cost come out to for a run? (Am not a startup.) Also, any smaller datasets you would recommend that still demonstrate some useful capability? Thanks.

Update: added result section to my post with more details: https://16x.engineer/2023/12/29/nanoGPT-azure-T4-ubuntu-guid... I don't think it can go any smaller and demonstrate its capabilities. Even GPT-2 is mostly generating nonsensical sentences that resembles English. You might be looking at ngram or markov model for simpler NLP capabilities.

Thanks!