Hugging Face

Hugging Face (Twitter) RT @karpathy: Last night I taught nanochat d32 how to count 'r' in strawberry (or similar variations). I thought this would be a good/fun example of how to add capabilities to nanochat and I wrote up a full guide here: https://github.com/karpathy/nanochat/discussions/164 This is done via a new synthetic task `SpellingBee` that generates examples of a user asking for this kind of a problem, and an ideal solution from an assistant. We then midtrain/SFT finetune on these to endow the LLM with the capability, or further train with RL to make it more robust. There are many details to get right especially at smaller model sizes and the guide steps through them. As a brief overview: - You have to ensure diversity in user prompts/queries - For small models like nanochat especially, you have to be really careful with the tokenization details to make the task easy for an LLM. In particular, you have to be careful with whitespace, and then you have to spread the... Перейти на оригинальный пост

Hugging Face (Twitter) RT @johntzwei: Announcing 🔭✨Hubble, a suite of open-source LLMs to advance the study of memorization! Pretrained models up to 8B params, with controlled insertion of texts (e.g., book passages, biographies, test sets, and more!) designed to emulate key memorization risks 🧵

Hugging Face (Twitter) RT @wjb_mattingly: Over the last 24 hours, I have finetuned three Qwen3-VL models (2B, 4B, and 8B) on the CATmuS dataset on @huggingface . The first version of the models are now available on the Small Models for GLAM organization with @vanstriendaniel ! (Link below). These are designed to work with line-level medieval languages and scripts represented in CATmuS

Hugging Face (Twitter) RT @arnicas: I made a little fairytale embeddings web app, using a tiny model from @huggingface entirely in the browser -- select text to navigate to a related fairytale text! find things! 1/2

Hugging Face (Twitter) RT @cerebras: Cerebras inference growth on @huggingface Reminds us of a certain Pink Floyd song..

Hugging Face (Twitter) RT @MaziyarPanahi: what happened this week with OCR and VLMs? * deepseek-ocr * chandra-ocr * nanonets-ocr2 * paddleocr-vl * qwen3-vl (2B, 32B, Instruct and Thinking) * dots.ocr * olmOCR 2 (based on Qwen2.5-VL) * LightOnOCR (smallies) top 5 trending models on @huggingface are still OCR/VLM!

Hugging Face (Twitter) RT @Thom_Wolf: Boom! More open-source collaboration between NVIDIA and Hugging Face robotics teams Releasing Gr00t N1.5 in LeRobot, an open foundation model by @NVIDIARobotics Try it, first feedback from the community is that “it’s a good model sir” https://twitter.com/LeRobotHF/status/1981334159801929947#m

Hugging Face (Twitter) RT @eliebakouch: pytorch RL env ❤️ hf hub https://twitter.com/_lewtun/status/1981380372748521929#m

Hugging Face (Twitter) RT @wjb_mattingly: Lot's of Qwen 3-VL finetunes on the way! I've got 6 GPUs going brrrr right now. These aren't the final versions of these models. Just the first tests.

Hugging Face (Twitter) RT @bhutanisanyam1: OpenEnvs for Reinforcement Learning! 🙏 We are launching a universal RL Environment interface today, teaming up with @huggingface and @UnslothAI Let’s take a trip down memory lane: It’s 2016, you read some papers. RL looks promising. But the reality? Cartpole is best we can train at home Fast forward to 2025, @danielhanchen teaches everyone how to train reasoning agents with free GPU Problem still remains How do you go beyond cartpole? You need access to good environments. OpenEnv defines a clean unified API. One interface. Any Environment. Zero Setup. It’s a huge honour to contribute with my heroes: @_lewtun, @lantiga and so many more Learn more: https://github.com/meta-pytorch/OpenEnv/

Hugging Face (Twitter) RT @_lewtun: Excited to share OpenEnv: frontier-grade RL environments for the open-source community 🔥! huggingface.co/blog/openenv 🧩 Modular interfaces: a clean Gymnasium-style API (reset(), step(), state()) that plugs into any RL framework 🐳 Built for scale: run environments in containers or servers with simple HTTP access for distributed training. 🌍 Frontier tooling in the open: developed by Meta & Hugging Face to bring robust RL infra to everyone. Next step: allow anyone to create, package, and share environments on the Hugging Face Hub 🤗 Let's go!

Hugging Face (Twitter) RT @HuggingPapers: Hugging Face just unveiled FineVision: The largest & cleanest open dataset for VLMs A meticulously curated corpus of 24 million samples, unifying 200+ sources into 185 subsets via a semi-automated, human-in-the-loop pipeline. Outperforms existing open mixtures, accelerating data-centric VLM research.