🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

This is a Windsurf stan account now Been using it for a month and have successfully vibe coded my way through final projects and everyday problems. Also student discount 🙇‍♂️

We're open-sourcing the training code for MetaMorph! MetaMorph offers a lightweight framework for turning LLMs into unified multimodal models: (multimodal) tokens -> transformers -> diffusion -> pixel! This is our best take on unified modeling as of November 2024, and

Am already a power user, reading documentation was a past era - now you can just ask for exactly the code flow you want to understand

the duality of man

Super random but I was at a themed cafe today and even o4-mini-high using VoT couldn’t solve it correctly after >10 min There’s a gap for a multilingual VLM reasoning eval here …to help my gf solve this by this time next year

TL;DR: Data can strongly change the power law exponent in scaling law, but tweaking architectures or optimizers rarely has the same impact. Very interesting observations.

claude code is just overall a quality of life upgrade - you can really just ask it to build all the little tools to improve the interfaces you interact with on the daily like go crazy with cli tools

We've improved our image-text curation significantly from our last blog post, now beating SigLIP2 through *data interventions alone* using vanilla CLIP. So proud of Ricardo Monti, Haoli Yin, Matthew Leavitt and the rest of the team! Check out the thread for all the details 👇

If you want to remain competitive, and ensure that your model improvements continue in the near and long term you MUST be investing in data curation. Very exciting to see these latest results from DatologyAI, which makes building better datasets suck far less.

The team absolutely crushed it here. They blew away nearly every CLIP baseline, and matched or exceeded SigLIP2—which uses a slew of training algorithm improvements—on a number of benchmarks. USING. DATA. CURATION. ONLY. I’m so proud of Ricardo Monti , Haoli Yin ,

We are back to show how adding a bit of magic to the training data alone can make CLIP outperform models that require a larger training budget and more sophisticated training algorithms.

Datology CLIP Models DatologyAI releases two SOTA CLIP ViT-B/32 variants: classification-optimized and retrieval-optimized, achieving top results through task-specific data curation alone. Model - ViT-B/32 (86M params), OpenCLIP 2.24.0 - No architecture or training changes -

Join DatologyAI if you have conviction on #4

Our customers needed a better base model <10B parameters. We spent the last 5 months building one. I'm delighted to share a preview of our first Arcee Foundation Model: AFM-4.5B-Preview.

Andrej Karpathy This is our exclusive focus DatologyAI. Data quality is the single most underinvested area of ML research relative to its impact. We've already been able to achieve 10x efficiency gains over open-source datasets, and I'm confident there's still another 100x because there's

It depends on how much you know about what you're using your model for. You want your data to be as similar to your test distribution as possible. In practice, benchmarks are an incomplete description of your true test distribution, so you want to hedge diversity vs.

the tokens - they seek to be learnable, not "good"