Kyunghyun Cho(@kchonyc) 's Twitter Profileg
Kyunghyun Cho

@kchonyc

a combination of a mediocre scientist, a mediocre manager, a mediocre advisor & a mediocre PC at @nyuniversity (@CILVRatNYU) & @genentech (@PrescientDesign).

ID:48642190

linkhttps://www.kyunghyuncho.me calendar_today19-06-2009 07:35:16

12,4K Tweets

60,6K Followers

2,2K Following

Yi Tay(@YiTayML) 's Twitter Profile Photo

It's been a wild ride. Just 20 of us, burning through thousands of H100s over the past months, we're glad to finally share this with the world! 💪

One of the goals we’ve had when starting Reka was to build cool innovative models at the frontier. Reaching GPT-4/Opus level was a…

account_circle
Kyunghyun Cho(@kchonyc) 's Twitter Profile Photo

i was suddenly reminded of this article from 2018.

'.. statistical physicists were primed to see power laws everywhere .. there’s a “power law religion.”'

'“We would even squint at the computer screen from an angle to get a better idea if a curve was straight or not,” recalled…

i was suddenly reminded of this article from 2018. '.. statistical physicists were primed to see power laws everywhere .. there’s a “power law religion.”' '“We would even squint at the computer screen from an angle to get a better idea if a curve was straight or not,” recalled…
account_circle
Kyunghyun Cho(@kchonyc) 's Twitter Profile Photo

oh i think i figured out:

1. create a new repository on github
2. clone the repository in Lightning AI ⚡️ studio
3. press 'Open' to work directly on the cloned repo

oh i think i figured out: 1. create a new repository on github 2. clone the repository in @LightningAI studio 3. press 'Open' to work directly on the cloned repo
account_circle
Kyunghyun Cho(@kchonyc) 's Twitter Profile Photo

so ... what's the right way to change the folder/repository name from 'this_studio' to something more informative in Lightning AI ⚡️ studio?

account_circle
Awni Hannun(@awnihannun) 's Twitter Profile Photo

Quantized Gemma 2B running at 157 toks/sec in MLX on my M1 Max laptop

Bunch of apps open (slack, chrome, etc) + attached display + screen recording!

Not sped up:

account_circle
Soumith Chintala(@soumithchintala) 's Twitter Profile Photo

Kyunghyun Cho PyTorch this is true, i can confirm. It wont stay this way, but I can attest that MLX and llama.cpp are moving much faster right now in developing for MacBook GPUs than PyTorch is.

account_circle
Kyunghyun Cho(@kchonyc) 's Twitter Profile Photo

... anyhow ... Lightning AI ⚡️ studio saved the day, and all i wanted to create this simple starter script for my friend ...

github.com/kyunghyuncho/q…

... anyhow ... @LightningAI studio saved the day, and all i wanted to create this simple starter script for my friend ... github.com/kyunghyuncho/q…
account_circle
Kyunghyun Cho(@kchonyc) 's Twitter Profile Photo

'Give a man a fish, and you feed him for a day. Teach a man to fish, and you feed him for a lifetime.'

- Prescient LM (2024)

'Give a man a fish, and you feed him for a day. Teach a man to fish, and you feed him for a lifetime.' - Prescient LM (2024)
account_circle
Kyunghyun Cho(@kchonyc) 's Twitter Profile Photo

of course, SFT on the positive responses from a binary preference data cannot work. if y_+, y_- \sim p_{\theta}(y|x), it's probably that y_+ and y_- are deemed similar to each other by the model(\theta), and thus p(y_+|x)↑ iff p(y_-|x)↑.

i heard it's called generalization

account_circle