23 articles
OpenAI trains a model using process supervision (rewarding each correct reasoning step) instead of outcome supervision (rewarding final answers). This approach achieves state-of-the-art mathematical problem solving and improves alignment by directly training models to produce human-endorsed chain-of-thought reasoning.
Hugging Face integrates BERTopic, a BERT-based topic modeling tool, directly into the Hub. Users can train, share, and deploy BERTopic models through Hugging Face's unified interface.
Hugging Face releases an optimized LLM inference container for Amazon SageMaker, enabling streamlined deployment of language models in production with improved performance and scalability.
OpenAI launches a program awarding ten $100,000 grants to fund experiments in democratic processes for deciding what rules AI systems should follow, within legal bounds.
Hugging Face and Intel optimize Stable Diffusion for Intel CPUs using NNCF and Optimum. 8-bit quantization reduces model size by 75% and accelerates inference 2-3x on CPU with no significant visual quality loss.
Hugging Face and Microsoft launch Hugging Face Model Catalog on Azure. This integration enables users to access open-source models directly through Microsoft's cloud platform, streamlining deployment and inference in production.
Hugging Face introduces 4-bit quantization with bitsandbytes and QLoRA to reduce LLM memory requirements. The technique enables fine-tuning 65B parameter models on a single 24GB GPU, making training accessible to more users.
Hugging Face introduces InstructPix2Pix, an instruction-tuning method for Stable Diffusion enabling image editing via text instructions. The model learns to apply specific edits (style changes, object modifications, color adjustments) from a source image and natural language instruction.
Hugging Face and IBM partner to integrate Hugging Face open-source models into watsonx.ai, IBM's enterprise AI platform. The collaboration gives developers access to language and vision models through a unified interface, with fine-tuning and production deployment support.
Safetensors passed independent security audit confirming its safety. The format becomes default standard for model storage on Hugging Face, gradually replacing legacy formats like pickle.
OpenAI calls for early thinking on superintelligence governance—future AI systems dramatically more capable than AGI. The company frames foundational debate on control frameworks needed before their emergence.
OpenAI launches ChatGPT app for iOS with conversation sync, voice input, and access to latest model improvements.
Hugging Face introduces Q8-Chat, a model optimized for Intel Xeon processors delivering efficient generative AI. The model reduces size while maintaining performance, enabling deployment on standard CPU infrastructure without GPUs.
BigCode built large-scale near-deduplication infrastructure to clean code data. The system identifies and removes near-duplicates across billions of files, improving training dataset quality for code models.
Hugging Face demonstrates how to deploy a ChatGPT-like chatbot on a single GPU using ROCm. The guide covers memory optimization and efficient inference for large language models on AMD hardware.
Hugging Face introduces RWKV, an RNN model combining transformer advantages: training parallelization and linear inference complexity. Hybrid architecture eliminates the quadratic attention bottleneck.
Hugging Face selected by French Data Protection Agency (CNIL) for enhanced support program. Recognition of the company's commitment to data protection and regulatory compliance in AI.
Hugging Face introduces Assisted Generation, a technique reducing text generation latency by using a fast draft model to validate tokens with a main model. Significant speed improvement without quality loss.
OpenAI uses GPT-4 to automatically generate explanations for neuron behavior in large language models and score those explanations. A dataset of these explanations and scores for every neuron in GPT-2 is released.
Hugging Face publishes a guide for building a coding assistant with StarCoder. The open-source model generates code and can be integrated into applications via APIs. The article details architecture and practical use cases.
Hugging Face explores the architecture and capabilities of text-to-video models. The article details current approaches, technical challenges, and practical applications of this emerging technology.
Hugging Face introduces StarCoder, a language model specialized in code generation. The model achieves performance comparable to proprietary solutions on standard coding benchmarks.
Hugging Face releases installation and usage guide for its Unity API, enabling game developers to integrate AI models directly into Unity projects.