🤖 As AI-generated content is shared in movies/TV/across the web, there's one simple low-hanging fruit 🍇 to help know what's real: Visible watermarks. With others @hf.co, I've made sure it's trivially easy to add this disclosure to images, video, chatbot text. See how: huggingface.co/blog/waterma...

I'm writing an article series about creating tensors from scratch in Rust. #tensors #machine-learning #ml #ai huggingface.co/blog/KeighBe...

AI doesn’t get your culture?❌ butchers your language? 😤 With FeeL – you can fix that🛠️🌍 💬 Talk to AI in your language ✏️ Correct its mistakes 👁‍🗨 Watch it improve The more we use it, the smarter it gets for everyone! 👉 Try it now: huggingface.co/spaces/feel-... 👶🤖📈 #ai #genAI #llm

How should AI tools be designed to support rather than replace workers? At the Reshaping Work conference, I led a roundtable exploring AI’s impact on labor. We published a blogpost on our key takeaways on responsible AI and the future of work w/ Franco Bastida 🔗 www.rsm.nl/discovery/20... 🧵👇

I've put together some of the handier tools for building courses and educational material on the @huggingface hub. These should bootstrap you projects with quizzes, friendly sized model, usefule datasets, and informative spaces. Let me know if you use or need more. https://buff.ly/42qyanw

Manic few days in open source AI, with game changing development all over the place. Here's a round up of the resources: Here's a thread on it all:

Teachers and Students! Here's a handy quiz app if you're preparing your own study material. TLDR, It's a quiz that uses a dataset to make questions and save answers.

If you need long context for RAG, tool use, agents, or just because, Nvidia released a new library to make it super simple. TLDR: You can get 128k context at 50% less memory 🐳 Here's a blog post on everything:

What happened yesterday in the Chinese AI community? 🚀 huggingface.co/posts/AdinaY...

Deepseek just dropped a frontier reasoning model on the hub. It's 685 billion parameters of bleeding edge performance on COMPLEX tasks. Who's considering this for synthetic datasets, distillation, or pruning?

Playing around with AI agents, and I reckon Gradio spaces on the hub make the perfect tools. - super easy to connect your agents to a bunch of useful tools and apps. - find a Space you like on Hugging Face Hub or make your own with Gradio. - link it up with smolagents. 🧵

We’re launching a FREE course on LLM Agents 🥳 📖 Learn what Agents are 🕵️ Build your own Agents using the latest libraries and tools. 🎓 Earn a certificate of completion to showcase your achievement. Enroll now 👉 huggingface.us17.list-manage.com/subscribe?u=...

Agents need tools and the Hugging Face hub is full of them. You can use Gradio spaces on the hub as agent tools. I created a short list that I tried out and made. Here's an overview 🧵

Great deep dive blog post on Agents, covering all the fundamentals from the ground up. @mmitchell.bsky.social @sashamtl.bsky.social @giadapistilli.com @evijit.io huggingface.co/blog/ethics-...

Free course on Agents by Hugging Face. We just added a chapter to smol course on agents. Naturally, using smolagents! The course cover these topics: - Code agents - Retrieval agents - Custom functional If you're building agent applications, this course should help.

If you're looking for real talk from experience, check out this blogpost on emissions from generative ai models: huggingface.co/blog/leaderb...

What does o3 mean for small open source models on real domains? I've been asking myself this question a lot the last week, and I think it sets up an interesting path forward. OpenAI, is calling o3 a new paradigm. If that's true, then these two developments illustrate how that paradigm plays out: 🧵

Synthetic Datasets are the focus of smol course this week! Synthetic datasets supercharge applying models to your own use case, because you can do stuff like this: 🧵 > >

People are flexing their end of year stats, so I made this app to show @hf.co hub stats in a tidy design! Thanks @jfcalvo.hf.co and @ameeelie.bsky.social for the feature!

Argilla annotation is free from code! You can now build and export your annotation projects, right from the comfort of your UI!

Vision Language Models have been the topic of the smol course this week. Which is a free and open course! If you're working with VLMs and/or up-skilling, check it out. There's loads of useful material, discussion, and feedback in the repo. 🧵

🙅‍♀️ No-code end-to-end example to train your model 1️⃣ Use the Synthetic Data Generator to create your custom dataset 2️⃣ Use AutoTrain to use the generated dataset and train your model Check it here: huggingface.co/blog/synthet...

Small open models are on a roll! Hugging Face just released research to get a 3 billion parameter Llama model to outperforms it's 70 billion parameter variant. 🧵

Lack of data is often the first challenge when you're building real AI systems for specific languages or use cases. If you're dealing with this, try out this dataset generator. It generates a custom dataset for your use case, which you can use to train a model for classification or chat.

[SATURDAY POST] In case you were disorientated by growing extra limbs or defying physics with your movements, here's a round up of what happened in AI this week. p.s. sora jokey 🧵

If you're using LLMs for use cases, this free course just got real! Smol course now has 4 chapters, and the most important was just released. 🌎 Chapter 4 shows you how to evaluate models on custom use cases. Check out smol course here: https://buff.ly/3ZCMKX2

Desperate to contribute to the development of Scots language AI. I've just contributed 16 examples to this dataset: data-is-better-together-fineweb-c.hf.space/share-your-p...

who's fine-tuning LLMs for reasoning? This dataset has been trending for a few weeks and there's a list of models trained on it. - It has SFT formatted reasoning sequences, like those in o1. - You could incorporate these into post training to boost reasoning abilities.

Learn local and private LLMs with Hugging Face 🤗 : Participation is free so join now! Even better, there are minimal GPU requirements and no paid services. Start with, Instruction Tuning, Preference Alignment, Parameter-efficient Fine-tuning. GitHub: https://buff.ly/3ZCMKX2

came across this example in agent-as-a-judge from Meta. It uses agent-as-a-judge to evaluate the effectiveness of a DevAI app. - It's based on an open dataset. - It's more accurate than LLM as a judge - It explains its evaluation based on preferences, and requirements. https://buff.ly/49tN6CQ

Quick update from week 1 of smol course. The community is taking the driving seat and using the material for their own projects. If you want to do the same, join in! 🧵

👐 Open Image Preferences is an Apache 2.0 licensed dataset for text-to-image generation by the @hf.co community. This dataset contains 10K text-to-image preference pairs across image generation categories, using different model families and prompt complexities. Blog: huggingface.co/blog/image-p...

[SATURDAY REVIEW] 🫖 In case you spent the week pushing millions of TB to hugginface hub and missed everything. It’s all about learning to finetune llms, more vision language models, the EU AI act, generating playable 3D environments. 🧵

Announcing Global-MMLU - an improved MMLU Open dataset with evaluation coverage across 42 languages. The result of months of work with the goal of advancing Multilingual LLM evaluation. Built together with the community and amazing collaborators at Cohere4AI, MILA, MIT, and many more.

Google dropped PaliGemma2. It's a vison language model starting at 2b parameters. Here's some notebooks finetuning it in: @merve.bsky.social transformers with quantized LoRA: https://buff.ly/3CXfH7s Jetha Chan brings keras LoRA https://buff.ly/4iBgprp & JAX https://buff.ly/4f5DLCv

Day 2 of smol course and the community is building something here. 👷 If you want to get involved, you can do this: - read (and star) the repo - check out our new discord channel - open a PR to submit an exercise on module 1 - open an issue to improve the course - review another submission 🧵

smol course Day 1 ✅. I learnt that people are hungry for models they can own. 📚 Material focused on instruction tuning. Split into chat templates and supervised fine tuning. There's more to this subject than this, but we're keeping things smol. ⏩ If you haven't already, try out module 1! 🧵

For anyone interested in fine-tuning or aligning LLMs, I’m running this free and open course called smol course. It’s not a big deal, it’s just smol. 🧵>>

We took those TRL notebooks from last week and made a page from them. So if you're upskilling on finetuning or aligning LLMs, and want examples from the community (like Maxime Labonne Philipp Schmid Sergio Paniego Blanco), check it out! bsky.app/profile/benb... >> huggingface.co/docs/trl/mai...

[SATURDAY THREAD] ☕️ 🧑‍🎓 In case you spent the week reading GDPR legislation and missed everything. It’s all about vision language models and image preference datasets. >> 🧵 Here are the models and datasets you can use in your projects.

Solid developer and user experience boost in this release of Argilla: - in the ui you can explore datasets with a lush overview - in python, installation is way easier. If you haven't already, this is a great release to take argilla for a test drive.

I wanted to share an easy to use notebook, that folk could use to explore and analyse LLM benchmark results in lighteval. Here it is: 🧵>>

The best path forward in AI requires technologists to be reflective/self-critical about how their work impacts society. Transparency helps this. Appreciate Bsky for flagging AI ethics &my colleague’s response. Let’s make informed consent a real thing. More later; Recommend: bsky.app/profile/cfie...

It's pretty sad to see the negative sentiment towards Hugging Face on this platform due to a dataset put by one of the employees. I want to write a small piece. 🧵 Hugging Face empowers everyone to use AI to create value and is against monopolization of AI it's a hosting platform above all.

Even the best intentions can mess things up, but Daniel has done the right thing now and reversed all his actions. He certainly doesn't deserve to be bullied for it. >> let's get back to building a fair and open solution to the problems big tech has caused.

I had to debug idefics3/smolVLM fine-tune because the model kept on yapping 🤠 plot twist: the processor and the chat template had a different EOS token compared to what base model had, which somehow sneaked into the code 🥸

My fav SmolVLM use case is personal document support. This is my workflow: 1. clone the demo repo 2. pass it a document I don't want to send to some rando API 3. use simple extractive questions to find key areas of the document 🧵 Here's the workflow in case you want to keep your data local

The community has labelled over 3000 image preferences in a few hours. One open source image preferences dataset coming right up!

Let’s make a generation of amazing open source image generation models from high quality data. The best image generation models train on human preferences. Unfortunately, many of these datasets are closed. Let’s change that! 🧵 we're building a community dataset and we need help reviewing!

TRL is a cornerstone of LLM post training and imo it's the default to learn. There are great alternatives like Unsloth, Axolotl, and AutoTrain. But if you want a daily drive that does experimentation to production, it's TRL. 🧵 these community notebooks guide you through TRL's core: