Hugging Face releases a comprehensive guide for pre-training BERT using Transformers and Habana Gaudi accelerators. The tutorial covers data preparation, model configuration, and performance optimization on specialized hardware.
Hugging Face publishes a deployment guide for Vision Transformer (ViT) model on Google's Vertex AI. The tutorial covers integration, configuration, and production deployment of a computer vision model on Google's cloud infrastructure.
Hugging Face publishes a deployment guide for Vision Transformer (ViT) model on Kubernetes using TensorFlow Serving. Covers containerization, inference configuration, and production orchestration.
OpenAI releases an improved content moderation tool free for API developers. The Moderation endpoint replaces the previous filter with enhanced detection capabilities.
Hugging Face releases a comprehensive guide for training and fine-tuning Sentence Transformers models. The tutorial covers fine-tuning techniques, embedding optimization, and integration with the Hugging Face platform to deploy high-performance text representation models.
Article on Proximal Policy Optimization (PPO), a foundational reinforcement learning algorithm used to train AI models. PPO improves stability and efficiency of reinforcement learning compared to earlier methods.
Hugging Face presents text generation optimization combining TensorFlow and XLA. The technique compiles models to accelerate inference without architecture changes. Results reproducible on standard benchmarks.
Hugging Face integrates TensorFlow Serving to deploy TensorFlow vision models directly on its platform. This integration streamlines production deployment and improves scalability for vision models.
OpenAI deploys a new technique so DALL·E 2 generates images of people that more accurately reflect global demographic diversity, reducing bias and improving safety.
OpenAI reports that over 3,000 artists from 118 countries are using DALL·E 2 in its research preview. These creators integrate the tool into their workflows and guide feature development decisions.
Hugging Face demonstrates a playlist generator using Sentence Transformers to encode music preferences as vectors. The system compares song embeddings and recommends similar tracks via vector search.
Hugging Face publishes a guide on policy gradients with PyTorch. The article covers reinforcement learning fundamentals and implements classical algorithms. Reproducible code and examples included.
Intel and Hugging Face partner to democratize machine learning hardware acceleration. The partnership aims to make optimization and inference tools more accessible to developers by integrating Intel technologies into the Hugging Face ecosystem.
Hugging Face releases an annotated guide to diffusion models, explaining the mathematical mechanisms and practical implementation of the progressive noise-addition process for image generation.
Cohere, OpenAI, and AI21 Labs release preliminary best practices for deploying language models, applicable to any organization developing or deploying LLMs.
OpenAI develops techniques enabling models to express uncertainty in natural language rather than confidence scores. The approach improves reliability and transparency of model responses.
Graphcore and Hugging Face launch a new lineup of transformers optimized for IPU (Intelligent Processing Units). These pre-trained models are ready for deployment on Graphcore infrastructure, streamlining production use.
Hugging Face introduces Pull Requests and Discussions features. These enable users to collaborate on models, datasets, and spaces, streamlining community contributions and iterative feedback.
Codex now powers 70 different applications across various use cases through the OpenAI API.
Hugging Face introduces TAPEX, an efficient pre-training method for table understanding models without real data. The technique uses data synthesis and reinforcement learning to improve performance on table-based question-answering tasks.
OpenAI expands DALL·E 2 access by adding 1,000 new users weekly. Current users have generated over 3 million images and contributed to improving safety processes.
Hugging Face launches a fellowship program to support researchers and developers working on open-source AI projects. The program provides resources, mentorship, and community visibility to selected participants.
fastai joins the Hugging Face Hub. The popular machine learning platform integrates its ecosystem with Hugging Face infrastructure to streamline model and dataset sharing.
Hugging Face releases guidance on PyTorch Fully Sharded Data Parallel (FSDP) for accelerating large model training. FSDP distributes model parameters across multiple GPUs, reducing per-device memory and improving scalability.
Hugging Face releases a guide for using the Transformers library on Habana Gaudi accelerators. The tutorial covers installation, configuration, and optimization of deep learning models on this specialized hardware infrastructure.
Hugging Face launches an education program providing free access to models, datasets, and tools for students and educators to learn machine learning and AI.
Hugging Face releases a report on CO2 emissions from the Hub and hosted models. The article details the environmental impact of model training and deployment, with metrics and recommendations to reduce carbon footprint.
OpenAI examines Goodhart's law in AI: when a metric becomes an optimization target, it ceases to be a valid measure. The challenge lies in measuring and optimizing difficult or costly objectives without degrading actual performance.
Habana Labs and Hugging Face partner to accelerate Transformer model training. The partnership aims to optimize performance on Habana's Gaudi processors and facilitate adoption by the open-source community.
Hugging Face introduces Decision Transformers, transformer models trained with reinforcement learning for autonomous decision-making. These models can be used for control and planning tasks without requiring custom code.
Hugging Face announces an AI Research Residency Program to develop emerging talent. The program provides immersion in open-source research projects and practical AI challenges.
Hugging Face publishes a guide to fine-tune a semantic segmentation model on a custom dataset. The method covers data preparation, fine-tuning, and model evaluation.
Hugging Face and AWS optimize BERT inference on AWS Inferentia. Benchmarks demonstrate significant acceleration and cost reduction for production deployments.
Hugging Face integrates image search into its datasets platform. Users can now index and query image collections using vector embeddings, enabling discovery and organization of large-scale visual data.
OpenAI shares lessons learned on language model safety and misuse prevention. The company outlines its current thinking to help other AI developers address these critical challenges in deployed models.
Hugging Face releases a guide to fine-tune Vision Transformer (ViT) for image classification tasks using the Transformers library. The tutorial covers loading pre-trained models, data preparation, and training on custom datasets.
Hugging Face releases guidance on using Wav2Vec2 for large audio files. The approach chunks files and processes segments in parallel, overcoming memory constraints in speech recognition models.
Hugging Face enhances Wav2Vec2 by integrating n-grams into the Transformers library. This optimization improves speech recognition model performance without major architectural changes.
Deployment guide for GPT-J 6B model on Amazon SageMaker using Hugging Face Transformers. Covers inference setup and integration of tools to serve the model in production.
Hugging Face integrates Active Learning into AutoNLP via Prodigy. The technique reduces annotation data volume by intelligently selecting the most informative examples. Streamlined workflow for iterative labeling and model training.