Hugging Face

Deploys CARLA simulator on HF Spaces for convenience.

Efficiently routes computation through experts in MoE models.

Improves tensor conversion scheduling for MoE models.

Introduces dynamic weight loading for efficient MoE model handling.

Skill for training AI models using Hugging Face Jobs.

Enhances GRPO with dense per-token reward signals for LLMs.

Build interactive web components with scoped CSS and JavaScript.

Mood-based movie recommendation engine using AI and MongoDB.

New voice feature enhancing Hugging Face's AI capabilities.

Distributes custom hardware kernels for AI applications.

Addition of new models like GPT-OSS and Chatterbox.

Separate library for tokenization across browsers and server-side.

New version with WebGPU runtime and modular class structure.

Users can submit evaluation results for models via PR.

Datasets can register as benchmarks with automatic score aggregation.

Allows community to report and aggregate AI benchmark scores.

AI model for video generation and editing applications.

AI model for image generation and editing tasks.

Isolate repositories to specific teams within an organization.

Scans commits for malware, secrets, and Pickle vulnerabilities.

AI model for image generation with 4-bit quantization.

Bilingual model for semantic sentence highlighting in documents.

Advanced Arabic language model with hybrid architecture and extended context.

Automatically selects the correct tokenizer class for models.

Redesigned tokenization architecture for improved modularity and transparency.