Hugging Face

Hi3DGen ?? High-fidelity 3D geometry generation from a single image by leveraging normal maps as an intermediate representation Play with the fantastic app on Hugging Face now: https://lnkd.in/g99NpV_y

Generate lifelike audio in real-time without a GPU! ?? Check out orpheus-cpp: a llama.cpp port of orpheus 3b text-to-speech model with built-in support for sync and async streaming. ?????? ?????????????? ??????????????-?????? ???????????? -?? ??????????????_?????? Project code: https://lnkd.in/ekPpN9mc

We just turned the humble dataframe into a superweapon?? dashboarding will never be the same!! ?? new Gradio Dataframe has: - multi-cell selection - column pinning - search + filtering - fullscreen mode - accessibility upgrades, and more

is your vision LM in prod even safe? ?? ShieldGemma 2 is the first ever safety model for multimodal vision LMs in production by Google DeepMind, came with Gemma 3 ?? I saw confusion around how to use it, so I put together a notebook and a demo, find it in the comments ??

is your vision LM in prod even safe? ?? ShieldGemma 2 is the first ever safety model for multimodal vision LMs in production by Google DeepMind, came with Gemma 3 ?? I saw confusion around how to use it, so I put together a notebook and a demo, find it in the comments ??

The Bonus Unit 2, "AI Agent Observability & Evaluation," is now live on our Hugging Face agents course! ?? You'll learn to: ?? Instrument agents with OpenTelemetry ?? Track token usage, latency & errors ?? Evaluate with LLM-as-a-judge ?? Benchmark with GSM8K ?? Check out the course here: https://lnkd.in/d2jiTx6j

Create Infinite Photographs of You with InfiniteYou-Flux! Flexible photo recreation that better preserves identity compared to current solutions like Pulid, IP Adapter, etc. ?? ?? Current full-performance bf16 model inference requires a peak VRAM of around 43 GB. You can build InfU on your own hardware: https://lnkd.in/g9dc_vVh Or Play for free on Hugging Face: https://lnkd.in/gzF7rikZ

?? Generate high-quality podcasts with the voices you want! MoonCast is an open sourced, multi-lingual, and zeroshot model. You just need to upload two sample voices, create a script, and that's it, run the model--You get a ?? notebooklm-like podcast. Model and App are released on Hugging Face: https://lnkd.in/gUk2EssP

?? Big news for GPU poors: thanks to Hyperbolic and Fireworks AI, you can run?DeepSeek AI's?new model using Hugging Face Inference Providers. What has changed since V3? Here's my quick home experiment ?? DeepSeek silently dropped an update to V3 yesterday. Benchmark results are available, showing significant improvements over V3. Still, it is always a good idea to run new models on data you care about and see more detailed, fine-grained results. Now that we can all run these new models from Day 0 with no GPUs required, I wanted to share my approach with an example I created this morning: 1. I got a sample from the LIMA dataset (containing high-quality general instructions). 2. Run the instructions with V3 and the new version V3-0324. 3. Define and run a simple judge with Llama3.3-70B to compare the model responses. 4. Push the dataset and pipeline so you can check and run similar experiments! (see first comment) 5. Extracted the results with Hugging Face Data Studio. Results summary - LIMA is not very challenging, but it is still interesting to see the differences between the two models. - A majority of Ties indicate that both models are close for this domain and task. - But still, V3-0324 consistently wins over V3 (33 times vs 6 times). As usual, the dataset, prompts, and pipeline are open-source (see first comment). What other experiments you'd like to see?

StarVector is a multimodal vision-language model for generating SVG (Scalable Vector Graphics). ?? It can be used to perform image2SVG and text2SVG generation. Live demo shows how the image generation is treated similar to a code generation task, using the power of StarVector multimodal VLM! ?? ?? Play with the app on Huggingface: https://lnkd.in/gCzdEbvj ?? If you want to build the model locally with a gradio app: https://lnkd.in/gDzCpdDN