Hugging Face covers Direct Preference Optimization (DPO) methods for LLM tuning. These techniques align models with human preferences without requiring a separate reward model, reducing computational complexity compared to traditional RLHF approaches.
OpenAI funded 10 teams worldwide to design ideas and tools for collective AI governance. The program shares innovations, learnings, and calls for researchers and engineers to continue this work.
Hugging Face enables free ComfyUI workflow execution via Gradio on Spaces. Users can deploy image generation pipelines without infrastructure costs.
Hugging Face releases a comprehensive guide for building custom leaderboards. Concrete example: Vectara's hallucination leaderboard, which evaluates models on their ability to generate factual responses without fabricating information.
Hugging Face releases unified LoRA training scripts to simplify model fine-tuning. Initiative to standardize fragmented existing approaches and improve accessibility to efficient fine-tuning.
OpenAI releases governance practices for agentic AI systems, covering monitoring, operational boundaries, and control mechanisms. The document proposes frameworks for safely deploying autonomous agents in production.
OpenAI and Axel Springer announce a partnership to integrate journalism into AI technologies. Axel Springer becomes the first global publishing house to sign such a deal with OpenAI.
Hugging Face explains Mixture of Experts (MoE) architecture: a mechanism where a router directs inputs to specialized experts instead of using all parameters. Reduces latency and increases model capacity without proportional computational overhead.
Hugging Face introduces SetFitABSA, a few-shot method for aspect-based sentiment analysis. SetFit enables efficient model training with minimal examples, without fine-tuning the base model weights.
Hugging Face introduces Prodigy-HF, a direct integration of the Prodigy annotation tool with its platform. This solution streamlines data labeling for model training directly within the Hugging Face ecosystem.
OpenAI establishes a Preparedness team and launches a challenge to address catastrophic risks from highly-capable AI systems and strengthen safety protocols.
OpenAI, Anthropic, Google, and Microsoft announce a new Executive Director of the Frontier Model Forum and establish a $10 million AI Safety Fund.
Hugging Face releases a tool to interactively explore datasets with one line of code. The feature integrates visualization, filtering, and search directly in the web interface.
Hugging Face releases a deployment guide for embedding models via Inference Endpoints. The platform enables serving vector representation models in production with automatic scaling and built-in monitoring.
Hugging Face identifies chat templates as a hidden source of performance degradation. Lack of standardization in conversation formats across models creates silent incompatibilities. The platform proposes a unified solution to normalize these templates and improve cross-model compatibility.
Hugging Face deploys AI Comic Factory through its Inference API. The tool generates comic strips from text prompts using vision and image generation models. Demonstrates serverless integration for AI applications.
Hugging Face releases a benchmark of Llama 2 on Amazon SageMaker. The study measures inference and fine-tuning performance of the model on AWS cloud infrastructure, providing latency and throughput metrics.
Hugging Face launches Inference for PROs, a professional inference offering with priority model access, dedicated support and guaranteed SLAs. Cloud service for deploying and serving LLM models in production.
OpenAI announces an open call for its Red Teaming Network, recruiting domain experts to identify vulnerabilities and improve the safety of its models.
Hugging Face launches an object detection leaderboard to benchmark and compare computer vision models' performance on standardized datasets.
Fetch reduces ML processing latency by 50% using Amazon SageMaker and Hugging Face. The optimization focuses on model inference and data pipeline acceleration for production applications.
OpenAI releases a teacher's guide for using ChatGPT in classrooms. It covers suggested prompts, how ChatGPT works and its limitations, the effectiveness of AI detectors, and bias issues.
Hugging Face deprecates Git password authentication. Users must migrate to personal access tokens or SSH keys before the deadline. This change enhances repository security.
OpenAI uses GPT-4 for content policy development and moderation decisions. The approach reduces human moderator workload, accelerates policy refinement feedback loops, and improves labeling consistency.
Hugging Face Hub is now available on AWS Marketplace. Users can access models and datasets through their existing AWS account with integrated billing.
Hugging Face and BentoML demonstrate deploying DeepFloyd IF (image generation model) using BentoML. Practical guide covering containerization, scalability, and integration with Hugging Face ecosystem for production.
Hugging Face enables streamlined MusicGen deployment through Inference Endpoints. Users can generate music without infrastructure management, with Meta model support and direct API integration.
Hugging Face explores fully homomorphic encryption (FHE) for large language models, enabling inference on encrypted data without decryption. Experimental approach aimed at protecting user privacy during LLM processing.
Hugging Face introduces Huggy Lingo, an ML system to improve language metadata on its Hub. The tool automatically detects languages in models and datasets, enriching catalog information to improve discoverability and interoperability.
OpenAI launches a new industry body to promote safe and responsible development of frontier AI systems. The initiative focuses on advancing AI safety research, identifying best practices and standards, and facilitating information sharing between policymakers and industry.
Hugging Face analyzes the EU AI Act's implications for open-source models. The article examines how compliance requirements apply to open models and proposes considerations for developers and regulators.
Hugging Face hosts an open source AI Game Jam bringing together developers and creators. The event produces games leveraging open source AI models (Llama, Mistral, etc.). Results demonstrate growing adoption of AI in indie game development.
OpenAI rolls out custom instructions for ChatGPT, enabling users to set preferences that the model will apply across all future conversations.
OpenAI partners with American Journalism Project in a $5+ million deal to explore how AI development can support local news and ensure local news organizations shape the future of this emerging technology.
Hugging Face releases a guide for fine-tuning Stable Diffusion models on Intel CPUs. The method leverages CPU-specific optimizations to reduce memory requirements and accelerate training without dedicated GPUs.
Transformers.js enables building AI-powered web games directly in the browser without backend servers. The library runs pre-trained transformer models in JavaScript, providing computer vision, natural language processing, and content generation for interactive games.
Hugging Face presents a web app generator using open-source ML models. The approach combines multiple models to transform text descriptions into functional HTML/CSS/JavaScript code, demonstrating the viability of open models for complex code generation.
Hugging Face optimizes BridgeTower, a vision-language model, on Habana Gaudi2 accelerators. The implementation improves inference and training performance for multimodal tasks.
Hugging Face announces major changes to its Open LLM Leaderboard: revision of evaluation criteria, improved transparency and reproducibility of benchmarks. The goal is to align rankings with real user needs and reduce score gaming.
Hugging Face responds to the U.S. NTIA's public consultation on AI accountability. The organization advocates for open standards, model transparency, and safeguards against misuse, while opposing regulatory barriers that would favor large corporations.