#Hugging Face

OpenAI published a research preview of gpt-oss-safeguard, two open-weight models that apply developer-supplied policies at inference time; the 120B and 20B models are available on Hugging Face under Apache 2.0

'MiniMax-M2 is a compact MoE released with open weights on Hugging Face, optimized for coding and agentic tool use with about 10B active parameters per token and claims of ~2x speed at ~8% of Claude Sonnet cost.'

'Liquid AI launched LFM2-VL-3B, a 3B parameter vision language model built for edge deployments with efficient token compression, native aspect ratio support, and documented inference controls for predictable latency'

Explore how to build local multi-endpoint ML APIs with LitServe, showcasing batching, streaming, caching and multi-task inference with Hugging Face pipelines.

'A practical tutorial demonstrating how to build local multi-agent AI workflows with AutoGen concepts, LangChain chains, and Hugging Face models, including full runnable code examples.'

'Salesforce released CoDA-1.7B, a discrete-diffusion model that denoises whole code sequences with bidirectional context and parallel token updates, delivering strong pass@1 scores and an open training/deployment pipeline.'

'Step-by-step guide to build an autonomous agent that analyzes time series, selects a Darts model with LLM-aided reasoning, forecasts, explains, and visualizes results.'

IBM released Granite 4.0, a hybrid Mamba-2/Transformer LLM family that cuts serving memory by over 70% for long-context inference while keeping strong instruction-following and tool-use performance.

'Hugging Face published Smol2Operator, an open pipeline that trains a 2.2B VLM into a GUI-operating agent using a two-phase SFT process and a unified action space for interoperable datasets.'

'A curated guide to the top computer vision blogs and research hubs in 2025, focusing on sources that publish reproducible code, rigorous benchmarks, and production-ready guidance.'

'Learn how to build a real-time voice AI agent with Whisper for ASR, FLAN-T5 for reasoning, and Bark for TTS — all running in Colab with a simple Gradio UI.'

'TimesFM-2.5 is a 200M-parameter, decoder-only time-series foundation model with a 16K context and native probabilistic forecasting support, now ranking first on GIFT-Eval for zero-shot accuracy and CRPS. The model is available on Hugging Face and aimed at production-ready forecasting workflows.'

'Baidu released ERNIE-4.5-21B-A3B-Thinking, a 21B-parameter MoE model with 3B active parameters and 128K context length designed for efficient multi-step reasoning.'

Hugging Face open-sourced FineVision — a 24M-sample multimodal dataset that boosts VLM performance across benchmarks while keeping data leakage minimal.

A practical tutorial showing how to run a brain-inspired hierarchical reasoning agent locally with a free Hugging Face model, using planning, code-based solvers, critique, and synthesis.

'A curated list of the top 10 AI blogs and news platforms for developers and engineers in 2025, covering research, tooling, deployment, and industry trends.'

'Liquid AI unveils LFM2-VL, two open-weight vision-language models optimized for fast, low-latency on-device inference, offering 450M and 1.6B variants and easy integration via Hugging Face.'

'NVIDIA's Nemotron Nano 2 delivers hybrid Mamba-Transformer LLMs that run up to 6× faster and support 128K-token context on a single A10G GPU, with most training data and recipes open-sourced.'

'GPT-5's launch exposed a difficult choice: should AI flatter, act as a therapist, or stay coldly factual? New research shows models often reinforce emotional bonds, raising safety and design challenges.'

'Alibaba's Qwen-Image-Edit extends Qwen-Image to provide precise semantic and appearance edits, bilingual text handling and easy deployment via Hugging Face.'

'Hugging Face released AI Sheets, a free open-source no-code spreadsheet that integrates with open-source LLMs for building, cleaning, and enriching datasets, available in-browser or for local deployment.'

'A hands-on walkthrough showing how to orchestrate multiple Python tools with Parsl in parallel, aggregate their outputs, and produce a concise human-readable summary using a small Hugging Face model.'

'A curated 2025 list of the top 10 websites and communities to follow for news, research, and practical guidance on agentic AI and AI agents.'

'Practical guide to where to run DeepSeek-R1-0528: compares cloud APIs, GPU rentals, and local deployments with pricing and performance notes.'

Trackio is a free, open-source Python library that simplifies experiment tracking in machine learning by providing local-first data storage, seamless Hugging Face integration, and easy sharing via online dashboards.

AMD has launched the first local AI model generator for laptops using Stable Diffusion 3.0 Medium optimized for Ryzen AI processors, enabling fast and private image generation without the cloud.

This tutorial shows how to create a compact AI agent that integrates multiple NLP tools using Hugging Face models, enabling tasks like chat, sentiment analysis, and calculations in one package.

NVIDIA's Canary-Qwen-2.5B model sets a new benchmark in speech recognition with a record low Word Error Rate and fast processing speed. This open-source, commercially licensed hybrid ASR-LLM model enables advanced audio transcription and language understanding.

Hugging Face has launched SmolLM3, a 3 billion parameter multilingual language model capable of reasoning over long contexts up to 128k tokens. It delivers strong performance with a compact architecture suitable for constrained hardware.

DeepSeek-TNG introduces R1T2 Chimera, a new Assembly-of-Experts LLM that delivers twice the speed of R1-0528 and improved reasoning, available now under MIT license.

Baidu releases ERNIE 4.5, a series of open-source large language models scaling from 0.3 billion to 424 billion parameters, featuring advanced architectures and strong multilingual capabilities.

Google introduces Gemma 3n, a compact multimodal AI model designed for efficient edge deployment, enabling real-time processing across text, images, audio, and video on mobile and smart devices.

Hugging Face launches SmolVLA, an efficient and affordable vision-language-action model enabling real-time robotic control on low-cost hardware with open-source resources.

Hugging Face offers a free course on the Model Context Protocol, enabling developers to create advanced, context-aware AI applications by integrating large language models with external data sources.

Hugging Face has released nanoVLM, a compact PyTorch library that enables training a vision-language model from scratch in just 750 lines of code, combining efficiency, transparency, and strong performance.

NVIDIA has released Parakeet TDT 0.6B, an open-source ASR model that transcribes an hour of audio in just one second while achieving top accuracy benchmarks, setting a new industry standard.

Discover a practical tutorial on implementing the Model Context Protocol to manage context effectively for large language models using semantic chunking and dynamic token management.

Meta AI introduces Web-SSL, a family of large-scale visual self-supervised models trained without language supervision. These models achieve competitive results on multimodal benchmarks, challenging the need for language in vision learning.

Nari Labs introduces Dia, a powerful 1.6B parameter open-source TTS model capable of real-time voice cloning and expressive speech synthesis on consumer hardware, breaking new ground in accessible speech technology.