Shital Shah's Avatar

Shital Shah

@sytelus.bsky.social

AI at Microsoft Research. If universe is an optimizer, what is its loss function? Code infra lead for Phi series of models. Some Open Source Projects: Airsim, TensorWatch, Archai, NanuGPT All opinions are my own.

230 Followers  |  740 Following  |  11 Posts  |  Joined: 24.11.2024  |  1.839

Latest posts by sytelus.bsky.social on Bluesky

More about phi-4:
bsky.app/profile/syte...

08.01.2025 15:50 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
microsoft/phi-4 Β· Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

HuggingFace Link:
huggingface.co/microsoft/ph...

08.01.2025 15:50 β€” πŸ‘ 1    πŸ” 1    πŸ’¬ 1    πŸ“Œ 0
Post image

We have been completely amazed by the response to phi-4 release. A lot of folks have been asking us for weight release. Few even uploaded bootlegged phi-4 weights on HuggingFace πŸ€·β€β™‚οΈ.

Well, wait no more. We are releasing today official phi-4 model on HuggingFace!

With MIT licence!!

08.01.2025 15:50 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

We hope you will have as much fun playing and creating with this tiny beast of a model as we had building it.

Happy holiday chilln' for ya allπŸŽ„πŸŽ.

13.12.2024 03:37 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
Phi-4 Technical Report We present phi-4, a 14-billion parameter language model developed with a training recipe that is centrally focused on data quality. Unlike most language models, where pre-training is based primarily o...

Paper: arxiv.org/abs/2412.08905

Model is available now to try in Azure: ai.azure.com/explore/mode...

What's better than tokens? Tokens with logits!!

13.12.2024 03:37 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

This is the work of super hard working team.

Of course, none of this would have been possible without support from our amazing leadership team
Ece Kamar, Peter Lee.

13.12.2024 03:37 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

Phi-4 achieves this by pushing the art of synthetic data even further to induce reasoning abilities along with new advancements in post-training.

If you have been using Llama 3.x etc for reasoning tasks, you owe it to yourself to try out Phi-4!

13.12.2024 03:37 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

Personally, since I started working on Phi project at its inception, this is my most favorite model that we have ever shipped.

Remember prompts that many frontier models including o1-preview struggled with? Phi-4 gave correct answer super fast!

13.12.2024 03:37 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

Are you ready for an early Christmas present from our team at Microsoft Research?

Introducing the most powerful smol model ever built in the world!

Welcome to Phi-4! πŸ‘‡

13.12.2024 03:37 β€” πŸ‘ 12    πŸ” 1    πŸ’¬ 1    πŸ“Œ 1
Search Jobs | Microsoft Careers

Apply here today:

jobs.careers.microsoft.com/global/en/jo...

This is a rare and unique opportunity to work on some very rewarding reasoning problems along side the members of Phi team and our product group partners.

Excited? Apply TODAY!!

We look forward to talk with you soon!

11.12.2024 00:26 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

Do you want to work on exciting AI and reasoning problems this summer?

We have an intern position just for you!

The internships for PhD students at Microsoft Research is one amazing experience and opportunity to work with world class researchers and engineers! πŸ‘‡

11.12.2024 00:26 β€” πŸ‘ 6    πŸ” 4    πŸ’¬ 1    πŸ“Œ 0

@sytelus is following 20 prominent accounts