Hugging Face (Twitter)

RT @Alibaba_Qwen: Introducing the compact, dense versions of Qwen3-VL — now available in 4B and 8B pairs, each with both Instruct and Thinking variants.

✅ Lower VRAM usage
✅ Full Qwen3-VL capabilities retained
✅ Strong performance across the board

Despite their size, they outperform models like Gemini 2.5 Flash Lite and GPT-5 Nano, and often beat them on benchmarks spanning STEM, VQA, OCR, video understanding, agent tasks, and more. In many cases, they even rival our flagship Qwen2.5-VL-72B from just six months ago!

Plus, FP8 versions are also available for efficient deployment.

Hugging Face: https://huggingface.co/collections/Qwen/qwen3-vl-68d2a7c1b8a8afce4ebd2dbe
ModelScope: https://modelscope.cn/collections/Qwen3-VL-5c7a94c8cb144b
Qwen3-VL-8B-Instruct API： https://modelstudio.console.alibabacloud.com/?tab=doc#/doc/?type=model&url=2840914_2&modelId=qwen3-vl-8b-instruct
Qwen3-VL-8B-Thinking API：...

Перейти на оригинальный пост

Hugging Face (Twitter)

RT @_fracapuano: should we release a comprehensive 70+ pages tutorial on robot learning, with hands-on code examples using @LeRobotHF and @huggingface? 🤗

‌Hugging Face (Twitter)

RT @HuggingPapers: ByteDance just released FaceCLIP on Hugging Face!

A new vision-language model specializing in understanding and generating diverse human faces.
Dive into the future of facial AI.

https://huggingface.co/ByteDance/FaceCLIP

Hugging Face (Twitter)

RT @osanseviero: Introducing... Cell2Sentence Scale 27B🤏

Based on Gemma, it's an open model that generated hypotheses about cancer cellular behavior. In collaboration with Yale, we confirmed the predictions with experimental validation in living cells

Super excited about this one 🤯

‌Hugging Face (Twitter)

RT @sundarpichai: The model + resources are now on HuggingFace and GitHub so researchers can keep building and experimenting. More details here: https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/

‌Hugging Face (Twitter)

RT @AziziShekoofeh: Explore the model and resources on Hugging Face:

https://huggingface.co/vandijklab/C2S-Scale-Gemma-2-27B

Hugging Face (Twitter)

RT @LeRobotHF: 🚀 New Feature in LeRobot : Dataset Editing Tools for LeRobotDataset !

We’ve introduced a set of powerful utilities to make working with LeRobot datasets easier and more flexible than ever !

✨ What’s new ✨
• 🗑️ Delete specific episodes from existing datasets
• ✂️ Split datasets by fractions or episode indices
• ➕ Add or remove features easily
• 🔗 Merge multiple datasets into one unified set
• 💻 New CLI: lerobot-edit-dataset for simple, configurable dataset editing

More on these new tools in the docs, link below 👇

Hugging Face (Twitter)

RT @LeRobotHF: Exciting things at LeRobot! 🤖

We’ve just integrated Meta-World: a benchmark for testing multi-task and generalization abilities in robotic manipulation: https://huggingface.co/docs/lerobot/en/metaworld

We’ve also cleaned up our environments and standardized on Gymnasium ≥ 1.x.x and MuJoCo ≥ 3.0.0 across the board.

Train your policy with the Meta-World dataset → https://huggingface.co/datasets/lerobot/metaworld_mt50

All in LeRobot! next is making environments as easy to load as datasets or models 🚀

‌Hugging Face (Twitter)

RT @ClementDelangue: Beautiful work from @sundarpichai @demishassabis and team with open weights on HF:

I'm so excited about the application of AI for biology and chemistry, especially in the open like this for all to benefit! https://twitter.com/sundarpichai/status/1978507110477332582#m

Hugging Face (Twitter)

RT @HuggingPapers: Facebook just dropped HoneyBee, a massive new dataset for vision-language reasoning, on Hugging Face!

It contains 2.5M high-quality examples with chain-of-thought solutions, pushing VLM performance to new SOTA.

Hugging Face (Twitter)

RT @shreyasgite: Data collection is a high-value task. Even the Chancellor of Germany has to do his part. Friedrich Merz with @LeRobotHF SO100.

Hugging Face (Twitter)

RT @Xianbao_QIAN: How far has embodied AI gone

Check out this first real world VLA manipulation evaluation, ran by tirelessly ARX, Franka, UR5 and Aloha arms

Table 30 are trivial for human but still very difficult for robots. pi0.5 is leading but still scores <50%

A long way to go! Link below:

Hugging Face (Twitter)

RT @Xianbao_QIAN: PaddleOCR-VL-0.9B is mind blowing and it supports 109 languages!

Check it out on HF demo:

Hugging Face (Twitter)

RT @reach_vb: BOOM: We've just re-launched HuggingChat v2 💬 - 115 open source models in a single interface is stronger than ChatGPT 🔥

Introducing: HuggingChat Omni 💫
> Select the best model for every prompt automatically 🚀
> Automatic model selection for your queries
> 115 models available across 15 providers including @GroqInc, @CerebrasSystems, @togethercompute, @novita_labs, and more

Powered by HF Inference Providers — access hundreds of AI models using only world-class inference providers

Omni uses a policy-based approach to model selection (after experimenting with different methods). Credits to @katanemo_ for their small routing model: katanemo/Arch-Router-1.5B

Coming next:
• MCP support with web search
• File support
• Omni routing selection improvements
• Customizable policies

Try it out today at hf[dot] co/chat 🤗

Hugging Face (Twitter)

RT @vanstriendaniel: Not enough people know about/use PRs for datasets on @huggingface. For many dynamic datasets, this can be a good workflow for versioning datasets and improving them over time.

Hugging Face (Twitter)

RT @LeRobotHF: 🚀 New in LeRobot: Multi-GPU training is now supported!

We’ve integrated 🤗 Accelerate into our training pipeline, making it simple to scale your experiments across multiple GPUs with just one command.

Whether you’re fine-tuning policies or running large-scale robot learning, LeRobot now handles distributed training easily.

👉 PR: https://github.com/huggingface/lerobot/pull/2154
Let’s accelerate robot learning together ⚙️🤖

‌Hugging Face (Twitter)

RT @HuggingPapers: ByteDance just released Sa2VA on Hugging Face.

This MLLM marries SAM2 with LLaVA for dense grounded understanding of images & videos,
offering SOTA performance in segmentation, grounding, and QA.

https://huggingface.co/ByteDance/Sa2VA-InternVL3-14B

‌Hugging Face (Twitter)

RT @abidlabs: Why did we build yet another experiment tracking library?

We built @TrackioApp because experiment tracking shouldn’t be complicated. Most tools are cloud-heavy, bloated, or hard to customize. Trackio is different: it’s lightweight, local-first, and free.

Run it on your machine, store logs in SQLite, visualize experiments instantly with a clean dashboard, or deploy online if you want. Embed dashboards anywhere, from blogs to internal docs. The API mirrors popular logging libraries, so you can switch without rewriting your code.

At under 5,000 lines of Python, Trackio is small, open-source, and designed for extensibility. Fork it, tweak it, add what matters to you. No limits, no lock-in, just fast, flexible experiment tracking for ML developers who want control.

Give it a star ⭐:

Hugging Face (Twitter)

RT @engineerrprompt: Some interesting insights on open models/repos

- 1 million new open-source AI repos landed on @huggingface in only 90 days.
- @nvidia , historically a hardware vendor, is now the single largest contributor of open AI models (Nemotron, Cosmos, Gr00t, BioNeMo, Canary).
- Chinese labs have moved from followers to co-leaders: Alibaba’s @Alibaba_Qwen , @deepseek_ai , Baidu, Tencent, MiniMax, Z.AI, @ByteDanceOSS , @Kimi_Moonshot and Zhipu all ship updates that rival or beat Western models on public leaderboards.
- DeepSeek alone has >100 k Hugging Face followers and is pushing iterative V3 drops.
- Fine-tuning is democratized—hundreds of LoRA adapters appear daily, letting individuals tune foundation models with only hundreds of samples.
- Europe’s footprint is shrinking: outside @MistralAI Magistral and Stability’s image models, almost no EU players are visible in the open-source explosion.
- Daily download counts for top repos now...

Перейти на оригинальный пост