Research
KV Caching from Scratch in nanoVLM Delivers a 38% Generation Speedup
Hugging Face Blog · Jun 4, 2025 · 2 min read
Implementing a fundamental optimization from scratch is often the best way to truly understand it. That's exactly what...
AI Research
Cutting-edge AI research papers, breakthroughs, and academic developments.
595 articles
Research
Stability AI’s audio model runs on Arm CPUs, generating WAVs in seconds
Hugging Face Blog · Jun 3, 2025 · 2 min read
A software engineer and music producer has demonstrated that you don’t need a GPU cluster to generate studio-ready audi...
Research
H Company drops open-source Holo1 VLMs that slash web automation costs to $0.13 per task
Hugging Face Blog · Jun 3, 2025 · 2 min read
H Company just released Holo1, an open-source family of action VLMs that makes web automation dramatically cheaper with...
Research
SmolVLA packs 30% faster robot learning into a 450M model you can run on a MacBook
Hugging Face Blog · Jun 3, 2025 · 2 min read
Robotics research has a scaling problem, and it's not the one you think. While labs with deep pockets train massive vis...
Research
TRL v0.18.0 lets you train and serve LLMs on the same GPUs, cutting idle time
Hugging Face Blog · Jun 3, 2025 · 2 min read
The latest release of Hugging Face's Transformer Reinforcement Learning (TRL) library, v0.18.0, solves a maddening inef...
Research
JSON plus Python beats raw code: structured agents gain 7 points on reasoning benchmarks
Hugging Face Blog · May 28, 2025 · 2 min read
The way AI agents take action is getting a quiet but important upgrade. For years, the default has been JSON function c...
Research
DeepSpeed ZeRO-3 clashing with Liger GRPO loss? Shape mismatch halts Qwen2.5 finetuning
Hugging Face Blog · May 25, 2025 · 2 min read
Another day, another obscure shape mismatch tearing through a perfectly good training run. This time, the culprit is a...
Research
Dell ships Llama 4 in under an hour, adds full on-prem AI app catalog
Hugging Face Blog · May 23, 2025 · 3 min read
Dell is making a serious play for the air-gapped enterprise, and they just cut the timeline from weeks to under an hour...
Research
Hugging Face's 'tiny-agents' Builds a Python AI Agent in Just 70 Lines of Code
Hugging Face Blog · May 23, 2025 · 2 min read
Hugging Face has a compelling argument for AI skeptics who think building agents requires complex frameworks: you can d...
Research
Falcon-H1 Packs 7B-Class Punch Into a 0.5B Model Using a Hybrid Attention-SSM Brain
Hugging Face Blog · May 21, 2025 · 2 min read
The Falcon-H1 family dropped today, and it’s a direct challenge to the idea that bigger models are the only path to bet...
Research
Falcon-Arabic 7B beats models 4x its size on Arabic benchmarks
Hugging Face Blog · May 21, 2025 · 2 min read
The Technology Innovation Institute (TII) just dropped Falcon-Arabic, and the benchmark numbers are frankly startling....
Research
Hugging Face drops nanoVLM: a 135M-param vision model you can train from scratch in pure PyTorch
Hugging Face Blog · May 21, 2025 · 2 min read
Hugging Face just released nanoVLM, a deliberately minimal toolkit for training vision-language models that feels like...
Research
New Diffusers quantization slashes Flux GPU memory from 31GB to under 18GB
Hugging Face Blog · May 21, 2025 · 2 min read
Hugging Face just made it much cheaper to run the massive Flux image model locally. A new pipeline-level quantization c...
Research
10,000+ Hugging Face models hit Azure AI Foundry with same-day release promise
Hugging Face Blog · May 19, 2025 · 2 min read
Two years ago, Microsoft and Hugging Face began a modest experiment: make open models easier to find and deploy on Azur...
Research
Falcon-Edge proves 1.58-bit models can match 16-bit rivals—and you can fine-tune them
Hugging Face Blog · May 15, 2025 · 2 min read
The Falcon-Edge series tackles the central headache of extreme model compression head-on: you usually have to pick betw...
Research
Transformers hits 300+ architectures as Hugging Face pushes for total AI ecosystem dominance
Hugging Face Blog · May 15, 2025 · 2 min read
The Transformers library isn't just growing. It's quietly becoming the central nervous system for the entire machine le...
Research
Kaggle now auto-generates Hugging Face model pages from your notebooks
Hugging Face Blog · May 14, 2025 · 2 min read
Kaggle just made the two biggest platforms in open-source AI a whole lot closer. Starting today, a new integration lets...
Research
Whisper on HF Endpoints hits 8x speed boost via vLLM with zero accuracy loss
Hugging Face Blog · May 13, 2025 · 2 min read
Hugging Face just dropped a community-powered ASR endpoint that's hard to ignore. The new Whisper deployment, built on...
Research
AI’s new Swiss Army knives: VLMs are now reasoning, talking, and running on phones
Hugging Face Blog · May 12, 2025 · 2 min read
The vision language model space has gotten weird — in the best way. When we last checked in on VLMs in April 2024, LLaV...
Research
Generalist Robots Are Stuck in the Lab — Here's the Data Problem They Can't Solve
Hugging Face Blog · May 11, 2025 · 2 min read
The biggest hurdle for generalist robots isn't shaky hands or slow processors. It's data. Specifically, the fact that a...
Research
Gradio now lets 1M+ devs turn any Python function into an LLM tool via MCP
Hugging Face Blog · Apr 30, 2025 · 2 min read
Gradio, the Python library used by over a million developers monthly to build ML interfaces, just made a quiet but sign...
Research
Qwen-3's chat template reveals how to toggle AI reasoning on and off
Hugging Face Blog · Apr 30, 2025 · 3 min read
The new Qwen-3 model from Alibaba's Qwen team ships with a chat template that's a quiet masterclass in practical AI des...
Research
Intel's AutoRound shrinks LLMs to 2-bit with only 37 minutes of GPU time
Hugging Face Blog · Apr 29, 2025 · 2 min read
Here's a number that makes other quantization researchers wince: 2.1x higher relative accuracy at 2-bit precision. That...
Research
Meta drops Llama Guard 4: a 12B safety model that runs on a single GPU
Hugging Face Blog · Apr 29, 2025 · 2 min read
Meta just released Llama Guard 4, and the headline feature isn't just about better detection — it's about practicality....