Hugging Face (Twitter)

RT @ClementDelangue: DeepSite crossed 10k likes and is now the third most popular space ever. Create a website with natural language for free or almost thanks to open-source AI models like Qwen3 Coder, Kimi K2, or Deepseek...

Mind-blowing how this whole topic of AI powered website creation has been exploding! https://huggingface.co/spaces/enzostvs/deepsite

Hugging Face (Twitter)

RT @charles_irl: ICYMI, open models for transcription are very good now. In just the last few months, we've gotten @nvidia Parakeet and Canary, @kyutai_labs STT, and @MistralAI Voxtral.

Running your own transcription at scale is now 100x faster and 100x cheaper than using a proprietary API.

Hugging Face (Twitter)

RT @lucataco93: You can now run Kontext LoRAs via Huggingface 🤗

https://huggingface.co/fofr/kontext-make-person-real

Hugging Face (Twitter)

RT @victormustar: New TTS bomb dropped on Hugging Face. It has multispeaker support and the output quality looks amazing 🤯

Hugging Face (Twitter)

RT @jetbrains: Not every developer task requires a general-purpose LLM.

We’re betting on specialized focal LLMs – smaller, faster, and focused.

Join @jetbrains and @huggingface for a livestream on how focal models like Mellum will shape the industry.

📅 July 29, 6 pm CET

👉 Save your spot: jb.gg/45n7t8

Hugging Face (Twitter)

RT @Alibaba_Qwen: 🚀 Introducing Qwen3-MT – our most powerful translation model yet!

Trained on trillions of multilingual tokens, it supports 92+ languages—covering 95%+ of the world’s population. 🌍✨

🔑 Why Qwen3-MT?
✅ Top-tier translation quality
✅ Customizable: terminology control, domain prompts, TM
✅ Ultra-fast & cost-effective: from $0.5/million tokens (MoE)
✅ Built for scale: low latency, high concurrency

Enhanced with reinforcement learning for unmatched fluency & accuracy.
Now available via the Qwen API – start breaking language barriers today! 💬🌐

Hugging Face Demo：https://huggingface.co/spaces/Qwen/Qwen3-MT-Demo
ModelScope Demo：https://modelscope.cn/studios/Qwen/Qwen3-MT-demo
API Doc：https://www.alibabacloud.com/help/en/model-studio/translation-abilities
Blog：https://qwenlm.github.io/blog/qwen-mt/

‌Hugging Face (Twitter)

RT @PyTorch: SmolLM3-3B-8da4w: With #TorchAO & optimum-executorch, quantizing and exporting for mobile is a breeze.

Now ready for on-device deployment with #ExecuTorch, running at 15 tokens/sec on Galaxy S22. 🔗 Model card with recipes + checkpoints: hubs.la/Q03yGyTN0
#EdgeAI #PyTorch

Hugging Face (Twitter)

RT @novita_labs: ⚡️Qwen3-235B-A22B-2507 supported by Novita, is also live on Hugging Face!

☑️ Function Call
☑️ Structured Output

Play with it 👇

Hugging Face (Twitter)

RT @pollenrobotics: New Unity package available: Reachy 2's digital twin!

- Gives immersive 3D experience through AR/VR
- Fully controllable via Reachy 2 stack
- Perfect for robotics courses & HRI research

Explore robotics without the physical robot!

https://github.com/pollen-robotics/Reachy2-UnityDigitalTwin

‌Hugging Face (Twitter)

RT @VentureBeat: Qwen3-Coder-480B-A35B-Instruct launches and it 'might be the best coding model yet'

Hugging Face (Twitter)

RT @ClementDelangue: I'm notorious for turning down 99% of the hundreds of requests every months to join calls (because I hate calls!). The @huggingface team saw an opportunity and bullied me in accepting to do a zoom call with users who upgrade to pro. I only caved under one strict condition: everyone gets crammed into a single chaotic group call. So... here we are: https://huggingface.co/pro?promo=zoom-clem

Please don’t upgrade. I still very much do not want to do Zoom calls 😂😂😂

Hugging Face (Twitter)

RT @MaziyarPanahi: Introducing the OpenMed Oncology NER App on @huggingface

🔬 Excited to share my beautiful @Gradio App specialized in extracting cancer, genetics, and oncology entities! 🚀

Spoiler alert, I used the new Qwen3-Coder-480B-A35B model! 🔥 (thanks @_akhaliq ❤️)

Try it now 👇

Hugging Face (Twitter)

RT @Gradio: We are making it easier and easier to build and share your MCP Servers with Gradio!

🔥 MCP is now selected by default (instead of python) when you click on the 'Use via API' page.
🧭 The chosen language now appears as a query parameter in the web address, making it accessible programmatically.

Hugging Face (Twitter)

RT @ClementDelangue: So cool to already see versions of the amazing hand in the wild just 10 days after release. The power of open-source communities for robotics!

Also, very different than 99% of robotics projects who bullshit with unrealistic futuristic projects! https://twitter.com/ClementDelangue/status/1944055485797167344#m

Hugging Face (Twitter)

RT @reach_vb: Qwen on a ROLL! Thinking model that beats Gemini 2.5 Pro, O4 mini AND DeepSeek R1 too 🔥

Hugging Face (Twitter)

RT @Xianbao_QIAN: Intern-S1, a new multimodal model from @intern_lm

- 235B MoE + 6B vision encoder
- 5T multimodal tokens & 2.5T scientific-domain tokens
- great model for AI4S research
- support tool calling capabilities

Model on @huggingface: https://huggingface.co/internlm/Intern-S1

Hugging Face (Twitter)

RT @RisingSayak: Fast LoRA inference for Flux with Diffusers and PEFT 🚨

There are great materials that demonstrate how to optimize inference for popular image generation models, such as Flux. However, very few cover how to serve LoRAs fast, despite LoRAs being an inseparable part of their adoption.

In our latest post, @BenjaminBossan and I show different techniques to optimize LoRA inference for the Flux family of models from @bfl_ml for image generation. Our recipe includes the use of:

1. torch.compile
2. Flash Attention 3 (when compatible)
3. Dynamic FP8 weight quantization (when compatible)
4. Hotswapping for avoiding recompilation during swapping new LoRAs 🤯

We have tested our recipe with Flux.1-Dev on both H100 and RTX 4090. We achieve at least a 2x speedup in either of the GPUs. We believe our recipe is grounded in the reality of how LoRA-based use cases are generally served. So, we hope this will be beneficial to the community 🤗

Even...

Перейти на оригинальный пост

Hugging Face (Twitter)

RT @reach_vb: NEW: GLM-4.5 & GLM-4.5-Air from @Zai_org - competitive w/ claude 4 opus and beats Gemini 2.5 Pro, MIT license🔥

> GLM-4.5: 355B total params, 32B active (MoE)
> GLM-4.5-Air: 106B total params, 12B active (MoE)
> "Thinking mode" (complex tasks) + "Non-thinking mode" (instant responses)
>128K context length + native function calling

Impressive benchmarks:
> AIME24: 91.0 (vs. Claude 4 Opus’s 75.7)
> MATH 500: 98.2 (vs. GPT-4.1’s 96.7)
> GPQA: 79.1 (vs. Gemini 2.5 Pro’s 84.4)
> SWE-bench Verified: 64.2 (vs. Claude 4 Sonnet’s 70.4)
> Terminal-Bench: 37.5 (vs. Claude 4 Opus’s 43.2)

> MoE - Loss-free balance routing + sigmoid gates
> Deeper, narrower - More layers, fewer experts (better reasoning).
> GQA: Partial RoPE + 96 attention heads
> 15T general + 7T code/reasoning tokens

Pretty solid model, looking forward for Inference providers to rise up and start serving this model! 🤗