Cody Blakeney
@code_star
Data Dawg @datologyai | Formerly Data Research Lead @DbrxMosaicAI | Visiting Researcher @ Facebook | Ph.D | #TXSTFOOTBALL fan | http://linktr.ee/code_star
You might like
We are looking for a post-training lead at @datologyai we have gpus, you can make them go brrrr
time for us to AllGather once again
Nvidia x Prime Intellect: Open-Source Model Builders & Scaling Meetup October 23 · 5:30-8 PM · San Francisco Join us luma.com/yndatvdq
Things are transpiring
Are there memes you associate with one of your friends? I have a few. Their go to memes. When I see the template used I immediately think of them and my heart is full.
I have no dog in this fight, but as an outsider to the long time RL community this is pretty funny.
It's funny how almost a decade later, the frontier labs are back at the same spot, building RL gyms
It's funny how almost a decade later, the frontier labs are back at the same spot, building RL gyms
Our reinforcement learning toolkit, OpenAI Gym, is now in public beta: gym.openai.com.
I love that he looks like he is doing a bit even while he is shredding.
I don’t think nearly enough people know that Tim Robinson was photographed for Thrasher magazine
If scene kids existed in the Bay Area would they call their band "Panic! at the Frisco"?
Me learning what mxfp8 is from this tweet.
to put this in real terms this probably is like going from a cost of ~$100k -> $500 in just 2 years from hardware, training techniques, and data to reach this capability.
crazy how far data and models have come in such a short time. I could probably train an MoE in a single day with 1 H100 node with a higher pass@1
There seems to be strange parallel worlds that exist in academic research. One side seems to have decided SFTing on reasoning traces is all you need for post-training small models. And one side thinks GRPO is all you need. I'm sort of baffled by the whole thing.
nanogpt pass@1 > 30% would go so hard
For the record we did the radar plots for the exact reason you are thinking. We were thinking about video games.
A lot of people were really mad at us when we made radar plots for the gauntlet lol if we had known how far people would take it we would have reconsidered
United States Trends
- 1. Good Saturday 23.4K posts
- 2. Chalobah 3,001 posts
- 3. #SaturdayVibes 3,318 posts
- 4. Emiru 11.1K posts
- 5. Ohtani 237K posts
- 6. Massie 36.7K posts
- 7. #saturdaymorning 1,625 posts
- 8. #dominATE_celebrATE 78K posts
- 9. Babe Ruth 3,897 posts
- 10. World Series 65K posts
- 11. Forest 75.5K posts
- 12. #NoKings 26.1K posts
- 13. #HeartofTaehyung 52.1K posts
- 14. Sam Harris 1,421 posts
- 15. George Santos 95.6K posts
- 16. FDV 5min 3,107 posts
- 17. TwitchCon 27.5K posts
- 18. AI Alert 8,828 posts
- 19. TOP CALL 9,798 posts
- 20. Beck 33.6K posts
You might like
-
Abhi Venigalla
@ml_hardware -
Tri Dao
@tri_dao -
Jonathan Frankle
@jefrankle -
Sam Havens
@sam_havens -
Jan Leike
@janleike -
Matthew Leavitt
@leavittron -
Sharon Y. Li
@SharonYixuanLi -
Ofir Press
@OfirPress -
Vitaliy Chiley
@vitaliychiley -
Mihir Patel
@mvpatel2000 -
Michael Carbin
@mcarbin -
Tom Goldstein
@tomgoldsteincs -
labml.ai
@labmlai -
Mostafa Dehghani
@m__dehghani -
rohan anil
@_arohan_
Something went wrong.
Something went wrong.