Releases

Transformers added support for running llama.cpp GGUF quantized models within the library’s model-definition and inference workflows.

Hugging Face released Tokenizers v1, a major milestone for its Rust-based tokenization stack.

NVIDIA entered a definitive agreement to acquire Hugging Face for $12.9303 billion. The companies said Hugging Face would retain its brand, platform, open-source commitments, and multi-cloud, multi-accelerator approach pending closing.

HF Jobs added exposed-port workflows for running private services such as OpenAI-compatible vLLM servers directly on Hugging Face compute.

Hugging Face launched mutable, Xet-backed S3-like Storage Buckets for checkpoints, logs, traces, and intermediate AI data.

Transformers v5 entered release candidate with simplified model definitions, standardized attention interfaces, and support for more than 400 architectures.

The core Hugging Face Hub Python client reached v1.0 with a redesigned hf CLI, httpx networking, and Xet-based file transfers.

Hugging Face discontinued HUGS model-deployment containers and directed users toward enterprise collections and other deployment paths.

The launch of HF Jobs established an on-demand compute layer alongside the Hub’s storage, collaboration, and inference products.

Hugging Face launched pay-per-use compute jobs that run scripts and containers on CPUs and GPUs directly through the hf CLI.

Hugging Face and Pollen Robotics launched Reachy Mini, a compact open-source desktop robot aimed at developers, researchers, and educators.

Hugging Face released a fully open 3B multilingual model with 128K context, switchable reasoning, tool use, and a disclosed training recipe.

Hugging Face and NVIDIA introduced a managed service for research organizations to access large GPU clusters for foundation-model training.

Hugging Face released SmolVLA, a 450M vision-language-action model trained for affordable, open robotic manipulation.

Hugging Face began broad deployment of Xet-based storage and transfers across Hub repositories, bringing deduplication and faster large-file operations.

LeRobot introduced the SO-101, a refined low-cost open robot arm for collecting demonstrations and training robotic policies.

Hugging Face acquired Pollen Robotics, maker of Reachy robots, to accelerate its open-source robotics strategy and connect physical hardware with the LeRobot ecosystem.

Hugging Face reported that Gradio had reached one million monthly users, reflecting its adoption for open AI demos and applications.

Hugging Face released SmolVLM2 at 256M, 500M, and 2.2B parameters with support for video understanding as well as images and text.

Hugging Face launched a unified inference layer that lets developers call models across multiple providers through one API and billing experience.

Hugging Face launched Open R1 to openly reproduce and extend the data, training, and evaluation pipeline behind reasoning models.

Hugging Face released a 500M vision-language model for low-memory multimodal inference.

Hugging Face released a 256M vision-language model small enough for highly constrained devices while supporting image-and-text tasks.

Text Generation Inference added a multi-backend architecture supporting vLLM and TensorRT-LLM alongside its native backend.

Hugging Face released smolagents, a lightweight open-source framework for tool-using and code-executing AI agents.

Answer.AI and Hugging Face released ModernBERT, a modernized encoder-only model family for retrieval and classification workloads.

Hugging Face released SmolVLM, a compact 2B vision-language model designed to run with very low memory.

Hugging Face detailed a new storage and transfer architecture using Xet’s content-defined chunking to accelerate large repository operations.

Hugging Face released the second generation of its 135M, 360M, and 1.7B compact language models with stronger instruction following.

Gradio 5 introduced redesigned application theming, stronger security defaults, and improved AI app development workflows.

Hugging Face launched HUGS, optimized zero-configuration deployment containers for open generative models.

The LeRobot community introduced the SO-100, a low-cost open robot arm design for data collection and imitation learning.

Accelerate reached its 1.0 release-candidate milestone after becoming a core distributed-compute layer in the Hugging Face ecosystem.

Hugging Face released IDEFICS3, a more efficient multimodal architecture that became the basis for the SmolVLM family.

Hugging Face acquired XetHub to bring content-defined chunking, deduplication, and high-performance storage to large AI repositories.

Hugging Face released compact fully open language models at 135M, 360M, and 1.7B parameter scales.

Hugging Face refreshed the Open LLM Leaderboard with a harder, contamination-resistant suite and new scoring methodology.

Hugging Face acquired Argilla, an open-source platform for human and synthetic data curation and model feedback.

Hugging Face released FineWeb-Edu, an educationally filtered subset that showed strong gains for small-model pretraining.

Hugging Face released FineWeb, a 15-trillion-token filtered and deduplicated English web dataset for language-model pretraining.

Microsoft and Hugging Face expanded their collaboration around open models, Azure AI, and enterprise deployment experiences.

Hugging Face launched LeRobot to bring open models, datasets, policies, and reproducible tooling to real-world robotics.

Hugging Face released IDEFICS2, an 8B visual-language model with stronger OCR, document, and visual reasoning capabilities.

BigCode released StarCoder2 models at 3B, 7B, and 15B scales in collaboration with ServiceNow and NVIDIA.

Hugging Face and Google Cloud announced a strategic partnership integrating the Hub with Google Cloud infrastructure, Vertex AI, and TPUs.

Pollen Robotics launched Reachy 2, an open-source humanoid robot platform later brought into Hugging Face through its acquisition of Pollen.

Hugging Face released a production-grade server for serving text embedding and reranking models.

Hugging Face H4 released an improved Zephyr 7B beta model trained on expanded public and synthetic preference data.

Hugging Face H4 released an aligned 7B chat model trained with distilled direct preference optimization.

Hugging Face released Candle, a minimalist Rust framework for fast and portable machine learning.

Hugging Face raised $235 million in a Series D led by Salesforce Ventures at a $4.5 billion post-money valuation, with participation from major cloud and semiconductor companies.

Hugging Face released IDEFICS, an 80B open-access visual-language model supporting sequences of images and text.

Hugging Face launched managed training and deployment offerings in AWS Marketplace, simplifying enterprise procurement and use on AWS.

NVIDIA and Hugging Face announced a partnership to connect Hub developers with NVIDIA DGX Cloud for training and tuning generative AI models.

Hugging Face introduced Collections for curating and sharing related models, datasets, Spaces, and papers.

Hugging Face launched a daily research-paper discovery feed connected to community profiles, models, datasets, and discussions.

Hugging Face launched an open and reproducible benchmark leaderboard for comparing community language models.

BigCode released StarCoder, a 15.5B-parameter code model trained on permissively licensed code from The Stack.

Hugging Face launched HuggingChat, an open-source conversational interface initially powered by an OpenAssistant model.

AWS and Hugging Face announced a strategic collaboration to make generative AI more accessible and optimize Transformers training and deployment on AWS infrastructure.

Hugging Face expanded its organization offering with team and enterprise plans for private repositories, access control, governance, and support.

Hugging Face released Parameter-Efficient Fine-Tuning tools for adapting large models with methods such as LoRA.

BigScience released BLOOMZ and mT0, instruction-finetuned multilingual models derived from BLOOM and the xP3 task collection.

The BigCode project released The Stack, a large permissively licensed source-code dataset with tools for governance and opt-out.

Hugging Face launched managed, autoscaling dedicated endpoints for deploying models from the Hub.

Hugging Face released Text Generation Inference, a production server optimized for large-language-model generation.

Diffusers added first-class support for Stable Diffusion, enabling local and hosted generation workflows around the newly released open model.

Hugging Face launched its enterprise platform around Private Hub, secure collaboration, governance, deployment, and organizational support.

Hugging Face launched Private Hub as a secure and compliant platform for organizations to collaborate on models, datasets, and applications.

Hugging Face released Transformer Reinforcement Learning tooling for post-training language models with supervised, preference, and reinforcement-learning methods.

Hugging Face launched Diffusers, an open-source library providing modular pipelines for diffusion-based generative models.

The BigScience workshop released BLOOM, a 176-billion-parameter multilingual language model trained on the Jean Zay supercomputer.

Hugging Face launched Evaluate as a unified framework for model and dataset evaluation.

Hugging Face raised $100 million in Series C funding led by Lux Capital at a reported $2 billion post-money valuation.

Hugging Face introduced browser previews and automatic processing for datasets hosted on the Hub.

Hugging Face introduced SafeTensors as a fast tensor serialization format designed to avoid arbitrary-code execution.

Hugging Face acquired Gradio, bringing its open-source interface framework and team into the company while keeping the project open source.

Hugging Face launched Spaces with native Gradio support, allowing community members to host interactive machine-learning demos on the Hub.

Hugging Face launched Optimum to optimize Transformers models for specialized hardware and inference runtimes.

Hugging Face launched AutoTrain to automate model training and fine-tuning from user-provided datasets.

Hugging Face introduced a paid individual plan with expanded private, storage, and compute capabilities on the Hub.

Hugging Face helped launch the BigScience open research workshop, coordinating an international collaboration to train a large multilingual language model.

Hugging Face launched Accelerate to let PyTorch code run across distributed hardware with minimal changes.

Hugging Face raised $40 million in Series B funding led by Addition to grow its open-source machine-learning ecosystem and hosted platform.

Transformers v4 began its release cycle with broad framework and architecture support and a standardized model API.

Hugging Face began its hosted inference platform with a serverless API for models on the Hub, later expanding into dedicated endpoints and provider routing.

Hugging Face introduced a hosted API for running models from the Hub without managing serving infrastructure.

Hugging Face launched hosted dataset repositories alongside its Datasets library, creating a shared catalog for reusable machine-learning data.

Hugging Face released a unified open-source library and catalog for loading, processing, streaming, and sharing datasets.

Hugging Face made model repositories a central Hub product for versioning, documenting, discovering, and reusing pretrained models.

Hugging Face expanded from an NLP library into a Git-based collaboration platform for sharing models and machine-learning artifacts.

Hugging Face introduced its high-performance Rust-backed Tokenizers library for production-scale text preprocessing.

Hugging Face raised $15 million in a Series A led by Lux Capital to expand its open-source natural-language-processing platform.

Hugging Face released DistilBERT, a smaller and faster distilled BERT model retaining most of the teacher model’s language-understanding performance.

Hugging Face renamed pytorch-transformers to Transformers and expanded the library’s framework and architecture support.

Gradio launched as an open-source Python framework for rapidly building browser interfaces around machine-learning models.

Hugging Face released an open-source PyTorch implementation and pretrained checkpoints for BERT, the project that evolved into the Transformers library.

Hugging Face raised a $4 million seed round backed by investors including A.Capital, Betaworks, SV Angel, and Thirty Five Ventures.

Hugging Face raised approximately $1.2 million in early angel funding to develop its conversational AI startup.

Clément Delangue, Julien Chaumond, and Thomas Wolf founded Hugging Face in New York, initially building a conversational AI companion before pivoting toward open machine learning.