AI BriefingHuggingFace Blog news
The latest AI and developer news from HuggingFace Blog, with short summaries.
Visit website →Knapsack-based Data Pipeline
This proposes a method for building a data pipeline that minimizes padding waste to maximize multimodal model training efficiency.
HuggingFace Unveils 3B Model SmolLM3
HuggingFace has unveiled SmolLM3, which delivers outstanding performance at the 3B scale and supports a 128k long context.
NeurIPS 2025 E2LM Competition Held
A NeurIPS 2025 competition is being held to build a new benchmark that evaluates scientific knowledge and reasoning ability in the early training stages of LLMs.
Sparse Embedding Model Training Guide
A detailed guide on training and fine-tuning Sparse Embedding models using the Sentence Transformers library.
NVIDIA Unveils Document-Specialized VLM
NVIDIA has released Llama Nemotron Nano, an 8B-scale VLM with enhanced document understanding and OCR performance.
Google Unveils Gemma 3n for On-Device Use
Google's multimodal model Gemma 3n has been released through various open-source libraries.
SGLang integrates Transformers backend
SGLang now supports the Transformers backend, enabling high-performance inference for any Transformers-compatible model.
FLUX.1-dev, QLoRA Training on Low-Spec GPUs
This introduces a method for efficiently fine-tuning the FLUX.1-dev model with under 10GB of VRAM using the diffusers library and QLoRA.
Hugging Face adds Groq inference support
Groq's LPU-based ultra-fast inference service is now directly available on the Hugging Face Hub.
vLLM Improves Long-Prompt Bottleneck Issue
This analyzes the queue-blocking problem caused by long prompts in vLLM's chunked-prefill strategy and presents an optimization approach.
Hugging Face adds support for Featherless AI inference
Hugging Face now supports Featherless AI, a serverless inference service, as a new Inference Provider.
Hugging Face Unveils Kernel Hub
Hugging Face has unveiled Kernel Hub, which lets developers easily load optimized compute kernels to boost model performance.
NVIDIA Unveils Isaac GR00T N1.5 and Provides Tutorial
Introduces how to fine-tune NVIDIA's robot foundation model Isaac GR00T N1.5 on the open-source LeRobot SO-101 robot arm.
HF-NVIDIA, GPU Cluster Service Launched
Hugging Face and NVIDIA have launched a GPU cluster rental service for researchers.
GUI Agent Evaluation Tool ScreenSuite Launched
ScreenSuite, an integrated benchmark suite for evaluating GUI agent performance from multiple angles, has been launched.
GUI Automation VLM 'Holo1' Released
Holo1, an open-source VLM family specialized for GUI automation, and the web agent Surfer-H have been released.
TRL Adds vLLM Co-location Support
TRL has integrated vLLM into the training process, reducing GPU resource waste and improving GRPO training efficiency.
HuggingFace Unveils Compact VLA Model 'SmolVLA'
SmolVLA, an open-source 450M-scale robotics VLA model that can run on consumer hardware, has been released.
Improving Performance with Structured CodeAgent
It presents a method of combining CodeAgent with structured JSON output to enhance the reasoning and execution reliability of agents.
Liger Kernel Cuts GRPO Memory Usage by 40%
Liger Kernel has been integrated into TRL, reducing memory usage by 40% during GRPO training.