Hugging Face (Twitter)

RT @ClementDelangue: The main breakthrough of GPT-5 was to route your messages between a couple of different models to give you the best, cheapest & fastest answer possible.

This is cool but imagine if you could do this not only for a couple of models but hundreds of them, big and small, fast and slow, in any language or specialized for any task - all at inference time. This is what we're introducing with HuggingChat Omni, powered by over 100 open-source models including gpt-oss, deepseek, qwen, kimi, smolLM, gemma, aya and many more already!

And this is just the beginning as there are over 2 millions open models not only for text but image, audio, video, biology, chemistry, time-series and more on @huggingface!

Hugging Face (Twitter)

RT @victormustar: Introducing: HuggingChat Omni 💫

Select the best model for every prompt automatically 🚀

- Automatic model selection for your queries
- 115 models available across 15 providers

Available now all Hugging Face users. 100% open source.

Hugging Face (Twitter)

RT @jadechoghari: Stay tuned with @NVIDIARobotics folks, we’re expanding @LeRobotHF’s sim capabilities! I can train & teleop my SO-101 from real → sim, drop custom assets, and collect data from home (or HF office). finally making progress on the robotics dataset problem. Project launches soon 👀

Hugging Face (Twitter)

RT @NVIDIAAIDev: This is what 3 million downloads looks like. 🥳

We owe a huge thank you to the AI community for making Llama Nemotron Nano VL 8B a favorite.

🤗 Try now on @huggingface: nvda.ws/4nWmwbV

Hugging Face (Twitter)

RT @MaziyarPanahi: the top two trending models on @huggingface are both for OCR!

document processing is a hot topic, kids! 😈

Hugging Face (Twitter)

RT @xeophon_: why does the chinese doordash continue to drop models

‌Hugging Face (Twitter)

RT @natolambert: Another roundup of the latest models with @xeophon_ !
Fun parts:
1. Methods for accurately monitoring HF 🤗downloads
2. GPT-OSS is mostly fixed and loved now
3. The perils of hybrid reasoning models
4. The continued degradation of open datasets
& usual surprises from China

‌Hugging Face (Twitter)

RT @interconnectsai: Latest open models (#15): It’s Qwen's world and we get to live in it, on CAISI's report, & GPT-OSS update
After a quiet month, Qwen is back in full force.
https://www.interconnects.ai/p/latest-open-models-15-its-qwens-world

Hugging Face (Twitter)

RT @ClementDelangue: I love the diversity of trending open datasets these days. There’s no excuse anymore not to train your own models!

- Fineweb and a shuffle of it by @karpathy
- Webscale-RL, a large-scale reinforcement learning dataset from @salesforce
- SVQ, an audio dataset from @Google
- Awesome chatgpt prompts with almost 10,000 likes by @fkadev
- A subset of the Math dataset by @DanHendrycks
- Nemotron personas by @nvidia
- An arabic dataset by @rightnowai_co
- A curated dataset of 1.5M+ @github repositories
- Toucan-1.5M, the largest fully synthetic tool-agent dataset
- A scientific paper dataset from @arxiv
- A cybersecurity dataset from @NIST by @ethanolivertroy

These are just the current trending amongst over half a million public datasets on @huggingface! hf.co/datasets

Hugging Face (Twitter)

RT @ClementDelangue: Btw the model router is the amazing open model Arch-Router-1.5B by @katanemo_ @salman_paracha: https://huggingface.co/katanemo/Arch-Router-1.5B https://twitter.com/ClementDelangue/status/1979230512343585279#m

Hugging Face (Twitter)

RT @alex_prompter: 🚨 Hugging Face & Oxford just dropped the playbook for robot intelligence.

It’s called LeRobot, and it’s basically the “PyTorch of robotics.”

End-to-end code. Real hardware. Generalist robot policies. All open source.

Here’s why this is huge:

• Robots can now learn from data like LLMs not just follow equations.
• They’re training on massive multimodal datasets (video + sensors + text).
• One model can control many robots from humanoids to arms to mobile bots.
• Built entirely in PyTorch + Hugging Face Hub.

We’ve had “foundation models” for text, code, and images.

Now comes the foundation model for motion.

This isn’t just robotics research it’s the beginning of robots that learn, reason, and adapt in the real world.

GitHub: github. com/huggingface/lerobot

Paper: arxiv. org/abs/2510.12403

‌Hugging Face (Twitter)

RT @reach_vb: Letsss gooo! DeepSeek just released a 3B OCR model on Hugging Face 🔥

Optimised to be token efficient AND scale ~200K+ pages/day on A100-40G

Same arch as DeepSeek VL2

Use it with Transformers, vLLM and more 🤗

https://huggingface.co/deepseek-ai/DeepSeek-OCR

Hugging Face (Twitter)

RT @ClementDelangue: I’m sure every founder believes their company is unique. But @huggingface does feel like a true “n of 1”. Is that founder bias, or fact?

Hugging Face (Twitter)

RT @xeophon_: new whale dropped https://twitter.com/xeophon_/status/1980140081290555726#m

Hugging Face (Twitter)

RT @_akhaliq: DeepSeek-OCR

Contexts Optical Compression

‌Hugging Face (Twitter)

RT @krea_ai: you can downloads the weights now from HuggingFace with Apache-2.0 license.

https://huggingface.co/krea/krea-realtime-video

‌Hugging Face (Twitter)

RT @AdinaYakup: DeepSeek-OCR is out 🔥

https://huggingface.co/deepseek-ai/DeepSeek-OCR

✨High-accuracy OCR - MIT license
✨Fast GPU inference (FlashAttention 2, BF16)
✨Docs > Markdown
✨Works with transformers

Hugging Face (Twitter)

RT @HKydlicek: We’re releasing the full FinePdfs source code — plus new datasets and models! 🚀

📚 Datasets:
• OCR-Annotations — 1.6k PDFs labeled for OCR need
• Gemma-LID-Annotation — 20k samples per language (annotated with Gemma3-27B)
🤖 Models:
• XGB-OCR — OCR classifier for PDFs

Hugging Face (Twitter)

RT @mervenoyann: DeepSeek-OCR is out! 🔥 my take ⤵️
> pretty insane it can parse and re-render charts in HTML
> it uses CLIP and SAM features concatenated, so better grounding
> very efficient per vision tokens/performance ratio
> covers 100 languages

Hugging Face (Twitter)

RT @RayFernando1337: This is the JPEG moment for AI.

Optical compression doesn't just make context cheaper. It makes AI memory architectures viable.

Training data bottlenecks? Solved.
- 200k pages/day on ONE GPU
- 33M pages/day on 20 nodes
- Every multimodal model is data-constrained. Not anymore.

Agent memory problem? Solved.
- The #1 blocker: agents forget
- Progressive compression = natural forgetting curve
- Agents can now run indefinitely without context collapse

RAG might be obsolete.
- Why chunk and retrieve if you can compress entire libraries into context?
- A 10,000-page corpus = 10M text tokens OR 1M vision tokens
- You just fit the whole thing in context

Multimodal training data generation: 10x more efficient
- If you're OpenAI/Anthropic/Google and you DON'T integrate this, you're 10x slower
- This is a Pareto improvement: better AND faster

Real-time AI becomes economically viable
- Live document analysis
- Streaming OCR for...

Перейти на оригинальный пост