A new series of Phi-3 mini, small & medium models (MIT license (3.8B – 7B – 14B) Long-context
Phi-3-mini-128k-instruct: https://
huggingface.co/microsoft/Phi-
3-mini-128k-instruct
…
Phi-3 small 128k: https://
huggingface.co/microsoft/Phi-
3-small-128k-instruct
…
Phi-3 medium 128k: https://
huggingface.co/microsoft/Phi-
3-medium-128k-instruct
… Multimodal:
Phi-3-vision-128k-instruct:
@thom_wolf
-
Microsoft Releases Phi-3 Series: 3.8B to 14B Models
By
–
-
Open AI as Common Good: Ownership and Transparency Over Closed Systems
By
–
like imagining a future where AI is something people can own, understand, tweak, build and improve together instead of a closed/black-box product full of secrets A common-good knowledge built on top of open-standard like the internet is or even science it self (no one privately
-
Moving beyond lonely futures: reimagining AI product inspiration
By
–
oh no but also, what about finding inspiration for AI products in other places than movies picturing futures full of loneliness? (As much as I enjoyed the aesthetic of the movie it’s still a sad tale)
-
Low-cost 3D-printed hardware enables safe hands-on learning
By
–
you pay it in reliability/strength but a thing i really love about low-cost 3d-printed hardware is how it can be so harmless by design (plastic made with low-cost motors). kids can just play and experiment with it without risks/supervision and that’s such an amazing way to learn pic.twitter.com/L4oqjCoB1t
— Thomas Wolf (@Thom_Wolf) 20 mai 2024you pay it in reliability/strength but a thing i really love about low-cost 3d-printed hardware is how it can be so harmless by design (plastic made with low-cost motors). kids can just play and experiment with it without risks/supervision and that’s such an amazing way to learn
-
Exploring Local Fablabs and Democratizing AI Innovation
By
–
Would be very cool imo – our goal is to explore with local Fablabs and non AI/robots-specialists very soon
-
Learning Teleoperation Skills for Robotic Systems
By
–
friday evening – time to learn a bit of teleop skills apparently pic.twitter.com/sDoZkAwWRc
— Thomas Wolf (@Thom_Wolf) 18 mai 2024friday evening – time to learn a bit of teleop skills apparently
-
New LLM Benchmark: Gathering Withheld Test Sets from Community
By
–
proposal: should we start gathering –from research teams and the community– new withheld test sets for a new LLM benchmark? infra/backend/leaderboard templates on the HF hub are super adapted for this so it would be very low effort to maintain it even over the long term
-
GPT-4o Uncanny Valley: Does Emotional AI Sound Too Human?
By
–
Is gpt-4o the first AI model falling in the uncanny valley because of sounding too emotional?
-
Developer Creates Beautiful ASCII Documentation for LeRobot Library
By
–
the amazing @asoare159 providing beautiful docstring ascii-art like every lib should have in https://
github.com/huggingface/le
robot
… – and yes he's also this awesome coder who will search in the "superscripts and subscripts" blocks of UTF-8 to find the right characters for this KL-divergence -
LeRobot Open Source Library Tops GitHub Trending Rankings
By
–
…yes ok i know the latest iPad ad is horrible, fully agree… but in other news the OSS lib « lerobot » is at the top of GitHub trending all lang, exciting!