We (xAI / Grok crew) will do a Meet & Greet opposite the main entrance of ICLR 2024 at 11am local time today. Come by to chat about research, engineering, and life at xAI!

👀
x.ai/blog/grok-1.5v

Grok is going multimodal!

It’s incredible to see how fast a small, focused team can move. Kudos to the amazing team @xAI that made this possible

x.ai/blog/grok-1.5v

x.ai/blog/grok-1.5

sharing some highlights from our recent paper: language models scale reliably with over-training and on downstream tasks!
arxiv: arxiv.org/abs/2403.08540

104 models, 11M to 7B parameters, varying numbers of tokens, 3 datasets, eval’d on 46 tasks: github.com/mlfoundations/…

1/11

As good a time as any to say I recently graduated and joined @xAI.

It’s going to be an exciting year, buckle up =)

New Resource: Foundation Model Development Cheatsheet for best practices

We compiled 250+ resources & tools for:
🔭 sourcing data
🔍 documenting & audits
🌴 environmental impact
☢️ risks & harms eval
🌍 release & monitoring

With experts from EleutherAI, Allen Institute for AI,…

Introducing time vectors! Time vectors are a simple way adapt LMs to new time periods; our results suggest that time is encoded in the weights of finetuned models. Led by my incredible undergrad mentee, Kai Nylund!
Paper: arxiv.org/abs/2312.13401
Code: github.com/KaiNylund/lm-w… /1

Another week, another bump in OSS CLIP model performance, 84.4% top-1 ZS Imagenet, solid retrieval. This time via 'Data Filtering Networks' dataset. OpenCLIP and Hugging Face transformers compatible weights on the hub huggingface.co/apple?search_m…

Awesome plot showing the progress in CLIP-like model training! As both a user and a researcher, there are a couple of caveats I personally feel worth pointing out.

Our best model, G/14 with 83.0% zero-shot ImageNet top-1 accuracy, is released at both OpenCLIP (github.com/mlfoundations/…) and CLIPA (github.com/UCSC-VLAA/CLIPA)

🚀Big updates to OpenCLIP! We now support over 100 pretrained models, and many other goodies. Check it out!

github.com/mlfoundations/…

At Adept we are open-sourcing a new Multi-Modal model with a dramatically simplified architecture and training procedure (more below)

If you are at #ICCV2023 and want to use the power of language to improve your zero-shot image classification accuracy, let’s talk! I’ll be at poster 156 in the south foyer at 2:30 today