Joey (e/λ)
@shxf0072
Followers
4K
Following
10K
Statuses
3K
I speak fluent Python and Sarcasm.
(anti-de sitter space)
Joined February 2019
@polynoamial @MillionInt <thinking> reward hacking opportunity @MillionInt i'll complete zelda for you we can share a reward </thinking>
0
0
6
@danielhanchen @UnslothAI sry, i wanted to help you push it faster, next time I'll use unsloth gen (more hackable:)
1
0
4
RT @danielhanchen: We managed to fit Llama 3.1 8B < 15GB with GRPO! Experience the R1 "aha moment" for free on Colab! Phi-4 14B also works…
0
288
0
agi at home fr
You can now reproduce DeepSeek-R1's reasoning on your own local device! Experience the "Aha" moment with just 7GB VRAM. Unsloth reduces GRPO training memory use by 80%. 15GB VRAM can transform Llama-3.1 (8B) & Phi-4 (14B) into reasoning models. Blog:
2
4
61
underrated crack anon, must follow
stepping into tpot arena thanks to the push by @shxf0072: what are people's favorite *high quality* deep learning compiler courses? there's not many. i'm hacking on singularity systems: zero to hero to address this gap in the ecosystem.
1
0
11
@jackminong idk nothing stoping us, bcs we can generate multiple rollout from given point in batch, if you want to do that with something like mario or atari it will be time consuming, with llm you just do batched gen
0
0
3