not yet but already got folks lookin at it ✌🏻
thx to tech crunch for putting it so concisely: “The secret is using less, but better quality, data. Instead of training on a library of billions of images that can’t possibly all be quality controlled, described, or deduplicated, Ai2 curated and annotated a set of just 600,000.”
AI2’s Molmo shows open source can meet, and beat, closed multimodal models
The common wisdom is that companies like Google, OpenAI, and Anthropic, with bottomless cash reserves and hundreds of top-tier researchers, are the only ones that can make state-of-the-art foundation model. But as one among them famously noted, they “have…
we released our open multimodal language model Molmo today 🥳 🍝 secret sauce? really really high quality set of image+text pairs, which we'll release openly 🕹️ try it out: molmo.allenai.orgmolmo.allenai.org/bloghuggingface.co/collections/...
it’d prolly be tough cuz there’s also the other regional CL confs personally I treat NAACL, EACL, AACL as interchangeable — just another option if miss ACL & EMNLP deadlines — but have def noticed way fewer Asian submissions to NAACL compared to non-regional confs
adding onto this google scholar issues aren’t limited to generated fake papers this study also investigates google scholar citation count manipulation by posting fake papers & paying services to do the same arxiv.org/abs/2402.04607 rlly need better alternatives
🐠NAACL name change vote should be goin out to ACL members, check email inbox for noreply@electionrunner.com 🐟Proposal is keep NAACL acronym but change from “North American Chapter" to "Nations of the Americas Chapter" 🐙Blog post explaining name change: naacl.org/posts/2024-0...
yea theres more to full picture when comparing MoE vs fully dense, direct comparisons are a bit hard, like for Gemma 2-3B for example, OLMoE should have faster inference since it’s only 1B active vs 3B active, but it also uses more memory since 7B total vs 3B total