Andreas Steiner

andreaspsteiner.bsky.social

19 February 2025 18:06

Looking for a small or medium sized VLM? PaliGemma 2 spans more than 150x of compute! Not sure yet if you want to invest the time 🪄finetuning🪄 on your data? Give it a try with our ready-to-use "mix" checkpoints: 🤗 huggingface.co/blog/paligem... 🎤 developers.googleblog.com/en/introduci...

Mehdi S. M. Sajjadi

msajjadi.com

11 January 2025 10:57

TRecViT: A Recurrent Video Transformer arxiv.org/abs/2412.14294 Causal, 3× fewer parameters, 12× less memory, 5× higher FLOPs than (non-causal) ViViT, matching / outperforming on Kinetics & SSv2 action recognition. Code and checkpoints out soon.

Mark Riedl

markriedl.bsky.social

22 December 2024 21:21

Six months ago someone put a for-loop around GPT-4o and got 50% on the ARC-AGI test set and 72% on a held-out training set redwoodresearch.substack.com/p/getting-50... Just sample 8000 times with beam search. o3 is probably a more principled search technique...

Alex Nichol

unixpickle.bsky.social

20 December 2024 23:17

Train a normalizing flow: loss curve is spiky and it diverges. Add more parameters to each layer: loss curve is smooth; no divergence. Kind of surprised the underparametrized model diverged instead of just *not learning*.

Alexander Kolesnikov

handle.invalid

20 December 2024 15:17

With some delay, JetFormer's *prequel* paper is finally out on arXiv: a radically simple ViT-based normalizing flow (NF) model that achieves SOTA results in its class. Jet is one of the key components of JetFormer, deserving a standalone report. Let's unpack: 🧵⬇️