Prompt: "A man with a cardboard sign near Times Square that says "Open-source is better than closed-source" We've got Llama 3.1 on the LLM side, but we now also got Flux on the Diffusion side. Open-source FTW! Probs to Black Forest Labs Try it yourself here: huggingface.co/spaces/black-f…

🔥 NEW SOTA ON Hugging Face 🔥 You can now use birefnet, a state of the art AI model for background removal 👀 kudos to Niels Rogge and Wauplin for the help with this release ❤️ huggingface.co/ZhengPeng7/BiR…

As a document AI enthousiast, I appreciate that Google is the only one providing a native API for document understanding. cloud.google.com/vertex-ai/gene… For PDF use cases that involve multiple pages, Gemini is currently a no-brainer (cause the others don’t provide an API yet)

Big announcement today Hugging Face: We now have a unified API for tool use across models from Mistral AI, AI at Meta, cohere, Nous Research and more! That means that you can reuse the same simplified, portable code to add tool capabilities to all of those models! 🧵

The Polars x HF Datasets integration is live ! 🎉 Load + Process Hugging Face Datasets at ⚡speed 😈 (Polars is written in Rust 🦀!) There are 200k public datasets already and it supports private repos too. Code + Docs in thread👇

Gotta love this grumpy dude making expectations on AI much more realistic This time, he takes Anthropic's demo of Claude-3.5-Sonnet's "agentic coding", which turns out to run terrible code 😂 youtube.com/watch?v=x0y1JW…

The Transformers library now also supports the Depth Anything v2 metric (also called "absolute") depth estimation models! These can tell you exactly in meters how far off each pixel is from the camera. Kudos to contributor "bt2513" for adding! huggingface.co/models?other=a…

Have to say, for any PDF use case where layout is important, the following script is goated. Gemini + few-shots enables me to generate perfect JSON given a new PDF. Gives the best results so far. Next I want to fine-tune an open model to do the same. Bottleneck will be memory

In case you're wondering which code Indie hackers like @levelsio and Danny Postma are making thousands of dollars on, it's this Thank me later github.com/huggingface/di…

wake up baby, a new shape-optimized SigLIP has dropped 💝✨

For anyone wondering which model this is, it’s Depth Anything v2 available in the Transformers library, contributed by yours truly huggingface.co/docs/transform…

The future is here folks! Having lots of fun with huggingface.co/spaces/Kwai-Ko…

More and more SOTA works are integrated with Hugging Face. This time, I collabed with the author of NeuFlow v2, a SOTA optical flow model for edge devices. Running at 20 FPS on 512x384 resolution images on an NVIDIA Jetson! Read more here: huggingface.co/papers/2408.10…

Hey! This is Hugging Face, are you using tools with open-source LLMs? Yell at me here! What's working, and what isn't? Was anything confusing? I will absolutely push fixes right here and now, but I need people to tell me what issues they're encountering

Using the new response prefill feature on Hugging Face to force Llama to answer me with increasingly stretched analogies