Майкрософт представила мультимодальную большую языковую модель
Она называется Kosmos-1 и может воспринимать как текст, так и картинки и аудио. Подробности можно почитать в препринте: https://arxiv.org/pdf/2302.14045.pdf
А это примеры результатов работы:
Подробнее
What's in this picture? Looks like a duck. That's not a duck. Then what's it? Looks more like a bunny. Why? It has bunny ears. Description of three toed woodpecker: It has black and white stripes throughout the body and a yellow crown. Description of downy woodpecker: It has white spots on its black wings and some red on its crown. Question: what is the name of the woodpecker in the picture? Downy Here are eight images: 4 (?) 4_ ♦ ®\± + © The following image is: /___ B___ C 3 B ® ♦ ♦
Completion Input Prompt Completion Input Prompt Question: Explain why this photo is funny? Answer: Question: Why did the little boy cry? Answer: The cat is wearing a Because his scooter mask that gives the broke cat a smile. (1) (2) TorchScale - A Library for Transformers at (Any) Scale 5 + 4 What is TorchScale? Question: The result is? Answer: A library that allows transformers to work efficiently and effectively at any scale. (5) (6) Question: What is the hairstyle of the blond called? Answer: Question: When will the movie be released? Answer: pony tail On June 27 (3) (4) Question: What is the heart rate in this picture? Answer: 57 bpm 10:10 on a large clock (7) (8)
Completion Input Prompt Completion Input Prompt An image of An image of Question: What coffee Question: What is inside shop is in the background? this canned good? Answer: Answer: a girl blowing out a a group of people Starbucks Corn candle on her birthday posing for a wedding cake. photo. (1) (2) (3) (4) a Texas Longhorns A Fine, Fine School pillow case. a parking sign that says Hope you have done it "Waiting? Please turn off your engine." (7) (5) (6) (8)
What's in this picture? sausage roll. How to cook it? Soak sausage roll in ketchup. Bake in oven for 15 min. Cut and enjoy. Can I put cheese in the dish? Sure. But make sure it is melted. What kind of animal is in the picture? What's in this picture? It's a zebra. A screenshot of the Windows 10 shutdown window Where do they live in? 1 would like to restart my computer. Which button should 1 click? In Africa. Press OK. Tell me a joke about zebras. Now 1 would not like to restart. What can 1 do? Zebras get in trouble for crossing the road. Click Cancel. (9) (10) (11)
нейронные сети,Microsoft,Майкрософт,kosmos-1
Еще на тему
Отличный комментарий!