I would be much more intrested in a piece on what you can train with this kind of rig, rather than the rig itself
Here is some additional journey apart from the rig. https://sabareesh.com/posts/llm-intro/
All You Need Is 4x 4090 GPUs to Train Your Own Model
11–20 of 125 posts
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#12Hey HN I am sharing my experience on how i pretrained my own LLM by building a ML rig at home
this is a decent birds eye view thanks, could you expand on this to show how long it took to produce... what model you produced? What did you produce? what did you train for.. the posts seems to suggest its for diffusion purposes?
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#13All you need is a 4x 4090 GPUs and a dedicated 30 amp circuit.
You sure? In my experiments with multi gpu inference, I couldn't get anywhere close to max theoretical power draw.
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#14Fun for a wealthy hobbyist, but if you want to do real work, you’re better off renting from Runpod. Good blog though.
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#15Interesting that DLSS 3 is mentioned as an advantage?
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#16All you need is 4x 4090 GPUs to Train Your Own Model -- and $12000 to buy them
I was contemplating between building rig vs using the cloud but for some reason I want to get hands on. So you can always rent them for a fraction of cost
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#17Fun for a wealthy hobbyist, but if you want to do real work, you’re better off renting from Runpod. Good blog though.
One of the motivation is to do several distillation, experimentation, research. But as you mentioned there are better ways to do this
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#18All you need is 4x 4090 GPUs to Train Your Own Model -- and $12000 to buy them
Well, one or the other, at any rate :-)
Re: All You Need Is 4x 4090 GPUs to Train Your Own Model
#19Hey HN I am sharing my experience on how i pretrained my own LLM by building a ML rig at home
this is a decent birds eye view thanks, could you expand on this to show how long it took to produce... what model you produced? What did you produce? what did you train for.. the posts seems to suggest its for diffusion purposes?
Here is some LLM models not diffusion though.
https://huggingface.co/sabareesh88/fw14k
this post has additional details
https://sabareesh.com/posts/llm-intro/