Presenting STOW at #CoRL2023 our latest work learning to segment and track previously unseen objects for robot stowing and fetching on cluttered shelves in warehouses. Check out our poster at Poster Sec. 6 on Thur 😄

Super excited about the announcement of Phi-2 from our team at MSR!!

Our internship job post on Evaluating & Understanding Foundation Models is out. Sharing here a list of open challenges our team is excited to explore together with future research interns. Application link: jobs.careers.microsoft.com/global/en/job/… Vibhav Vineet Neel Joshi @hmd_palangi Ece Kamar

we have trained a humanoid transformer with large-scale reinforcement learning in simulation and deployed it to the real world zero-shot

Today, we share our teams’ latest contributions, Phi-2 and promptbase. Phi-2 outperforms other existing small language models, yet it’s small enough to run on a laptop or mobile device. msft.it/6040ipYH6

Excited to share the latest progress of phi-2 from our group! Check out the blog post for more details! 👇👇👇

Microsoft releases Phi-2 on Hugging Face model: huggingface.co/microsoft/phi-2 a 2.7 billion-parameter language model that demonstrates outstanding reasoning and language understanding capabilities, showcasing state-of-the-art performance among base language models with less than 13

Phi-2 by Microsoft AI is now the #1 trending model on Hugging Face (hf.co/models). 2024 will be the year of smoll AI models!

Check out this short video for a brief discussion of the phi series with Microsoft CTO Kevin Scott , including why "textbooks" in "Textbooks Are All You Need" might not be exactly what you have in mind. youtu.be/O-DjHgZt-Uk?si…

We're so pumped to see phi-2 at the top of trending models on Hugging Face ! It's sibling phi-1.5 has already half a million downloads. Can't wait to see the mechanistic interpretability works that will come out of this & their impact on all the important LLM research questions!

Are larger vision models always necessary? We find scaling on **image scales** (e.g., 224->448->672) is usually better than scaling on model size (e.g., Base->Large->Giant). With one line of code, improve any vision model for Multimodal LLMs or various vision and robotic tasks!

Thank you Aran for sharing our work!

We are releasing phi-3 mini today! Finally, we have an open-sourced SLM (3.8B) at GPT-3.5 level! Checkout the models and technical report at here huggingface.co/microsoft/Phi-… 🥳

S2 is officially integrated into NVIIDA VILA! Checkpoint for VILA-3b with S2 is released. More checkpoints on the way! S2 enables any vision model to perceive higher resolution with as few as **one line of code**. Try it out here: github.com/bfshi/scaling_…

Amazing work on these new benchmarks, keep them coming!!! And notice our little phi-3-mini (3.8B) ahead of 34B models :-). Quite curious to see where phi-3-medium (14B) lands!

Updated phi-3 tech report with final numbers for 7B/14B and a new section on phi-3-V (e.g., MMMU at 40.4, in the ballpark of Claude 3-haiku and Gemini-1.0 pro) : arxiv.org/abs/2404.14219

Great work Shishir Patil Tianjun Zhang It still feels like yesterday when we kicked out the project. Great to see the work continue to influence the function calling space 😉

Surprise #NeurIPS2024 drop for y'all: phi-4 available open weights and with amazing results!!! Tl;dr: phi-4 is in Llama 3.3-70B category (win some lose some) with 5x fewer parameters, and notably outperforms on pure reasoning like GPQA (56%) and MATH (80%).

The models are high again. We bring to you, o3 & o4-mini, our absolute best text & VISUAL reasoning models that truly manage to use any tools to solve hard tasks: canvas, browser, python, memory, ... & IMAGEGEN. The secret trick is to talk to the models in images🤫