How much compute does this require? I don't want to make a big llm

Really amazing 🎉

Hi, in my opinion, instead of using print() to debug, using breakpoint() or import pdb; pdb.set_trace() would be more easier to debug your code! try it out :)

You make a valid point about the knowledge acquisition gap in general. What concerns me about how we are discussing and socializing AI is the leap from that valid point to cognition & practice. From knowledge to practicing doctor/medicine is an enormous leap. AI no doubt can be an important part of this, but the parallels (both direct and indirect) drawn here about time in med school/residency vs what AI could do fundamentally discounts that med school (or any other type of training) is actually far less about knowledge acquisition as it is about practice. I know what it takes to run a 4 minute mile. But can I? I wish. I don’t think you were making that kind of a claim but I think it risks coming across that way.

🔥

Good stuff! 👏

I look forward to more of the content. but as an audio engineer... I can not stress how important it is for good audio on KZbin. You will lose so many people in seconds just because you didn't do EQ.

AI Vocal Trance 😄👍

I love to see young men hustling! #Kaizen, Your desire to learn and teach is invaluable! #LLM #Architecture #Future

the input_dim and output_dim must be the same, otherwise, the program failed.

You need 2 more screens

Great Work

subtitles are showing but voice is not coming only music is audible

Cool

Nice

Great content, get well soon @KyeGomez Thanks for sharing. Can you also please write code for training and a simple example to show its effectiveness.

Hope you get better soon

thanks. get well

Intro could have been Elevenlabs api call in Sam Elliott voice. Great Idea and execution. more vids.

happy with yet again, another super channel! 👏👏👏

"Taxes and weird stuff like that" hahahah

problem is that they don't work

Just heard about the mamba model and was trying to build a simple one from scratch. Thank you for your video and do you think these can be used for other tasks beside nlp?

Nice, setup a web page to use that online, because i dont want to spend time trying to setup or update that, charge for use, give a first 5$ free to try, use opensource LLMs, and optional corporate ai models, let the people use it. Dont use any user data for training, privacy is very important, make the note to the user that corporate ai is not private.

Yeahhhhhhhh

Why not incorporate something like Actor Model, kzbin.info/www/bejne/e33adpeYlNV3Zs0si=B4tK4_-w2uTGXNnn , that will bring fault tolerance. Thanks Kye for make your videos more short, we dont have time to see hour long videos.

discord link expired, u need to generate the without expiry one

dspy in the hishoussee!!

In general a good video. Learned how it is like when experimenting with a new architecture.

this random zoom effect is really annoying.

Bzzt. GQA was used in LLaMA-2, so it's not "in the last six months or so."

3:02 "A query is like a Google search, a value is like a web page, and a key is like the title." Bruh. I was a Senior Software Engineer at Google for over 11 years. What you are saying makes literally no sense.

Dude what happened to your amazing hair it was so original😭

Thanks @kyegomez3242. Great summary description.

Impressive model, thanks for the video.

incredibly based

Can't wait for the next videos in the series!

why is the voice so corrupted?

Your screen changes are too fast and not feeling stable and watchable.

TIME TO ACCELERATE 🚀

This is great! Cannot wait to try this out on my 3090.

Huge

Cool.

I've been working on something similar and the real problem with using LLM output as input into any kind of education is the grounding. There's little point in creating a curriculum that has hallucinations in it. It seems to me that one companion to a swarm would be a fine-tuning capability with accompanying data source that ensures that the output is grounded in fact. Maybe one way to do this could be to have a list of URLs of websites to crawl to feed into a fine tuning task, after which the curriculum is generated. Maybe the accompanying with a domain-specific HIT that validates output.

Is the language model learning to improve and if so is the policy dictated?

With the right models this could be big

I think the first thing to do when speaking to another person to find out if they are truly paying attention, is to randomly sprinkle 'multimodal tree of thoughts' every once in a while and nonchalantly continue. Just say it every once in a while, it's like a mini vitamin for your soul

I guess I'm still learning things. With all due respect am I the only one asking, what the hell is this? I need a cliff notes, I don't know what I'm looking at and what it's for. I missed the intro. This is a foreign world

That description text was written by AI.

Just subscribed! Keep up the good work mate