R to @YiTayML: Pretty much your original recipe too!

R to @giffmana: 👀👀👀 haha @giffmana I was wondering about the 3b ul2 when I first saw this paper. I left a lot ckpts lying around but I figured you guys trained a new one. 😃

R to @mervenoyann: @readwise save thread

R to @OpenIndexAI: Unfortunate that streaming support on @replicate is still quite limited

R to @OpenIndexAI: Model explosion!

Want to play with @AdeptAILabs's Fuyu-8B multimodal model? Try it out at https://www.openindex.ai/adeptfuyu Thanks to @lucataco93's model on @replicate

R to @karpathy: Feels like a feature @AdeptAILabs would be able to enable quickly

R to @Vaishaal: Yeah probably, or human eval, or multiple choice scoring, or make questions which only have one answer (ocrvqa). The multiple choice is almost what you mean, but not what people do when they classify vqav2. coco etal only make sense with fine tuning (and were designed for this)

R to @giffmana: So if you don't like zs generative vqa, zs classification vqa, so is the only way to evaluate vqa via finetuning? What about few shot vqa -- this should at least get around the verbosity problem for VLMs.

RT by @tendyman69420: 🧶PaLI-3 achieves SOTA across many vision-language (and video!) tasks while being 10x smaller than its predecessor PaLI-X. At only 5B parameters, it's also smaller (and stronger) than the concurrent Fuyu-8B model, though sadly we cannot release the model (props to @AdeptAILabs)

RT by @bansal22luvi: The multimodal model Fuyu-8b from @AdeptAILabs was released this week on @huggingface, and it's fully compatible with 4-bits quantization! Try it out yourself on this colab notebook: https://colab.research.google.com/drive/1cVfYTYNQkiXughf1iIS5T-ZlX6gt4yrH?usp=sharing h/t @younesbelkada for this!

RT by @llmgen_AI: Fuyu-8B released by @AdeptAILabs is a very impressive multimodal model capable of.. 🖼️ Visual Question Answering 🖼️ Image Captioning 🖼️ Text localization and more! I tested it and here are my findings 🧶 Also link to the Space 👉 https://hf.co/spaces/adept/fuyu-8b-demo

RT by @__MLT__: Wonderful work from @mervenoyann, now available to use in our public demo! https://huggingface.co/spaces/adept/fuyu-8b-demo

RT by @JayaGup10: We think MM models are especially useful for handling unstructured knowledge worker data, so we’ve given Fuyu-8B capabilities in: - Understanding diagrams, charts, and graphs - Doing OCR on screens - Outputting bounding boxes for the locations of objects on screens - Answering UI-based questions

RT by @Tiancaixinxin: I made a chatbot that controls browser by writing Selenium code at @scale_AI #GenerativeAI hackathon with @Tiancaixinxin. Thanks for prompt advice from @goodside. So glad I met @karpathy. It's similar to @natfriedman @AdeptAILabs @BrowseAI. blog and code: https://yihui.dev/actgpt