14 articles
OpenAI presents an efficient training method for language models capable of filling in text in the middle of sequences. The technique reduces computational costs while maintaining performance on standard tasks.
Hugging Face expands Datasets documentation with dedicated guides for audio and vision. New resources for integrating and processing images and audio files in ML pipelines.
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.
OpenAI releases a hazard analysis framework for code synthesis LLMs. The work assesses potential dangers of code-generating models, proposes evaluation methods and mitigation measures to improve safety.
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.
Article on Advantage Actor Critic (A2C), a reinforcement learning algorithm combining actor-critic approaches. Explains fundamental principles, architecture, and practical applications of this method.
DALL·E enters beta with access rolling out to 1 million waitlist users over coming weeks. Users get monthly refillable free credits and can purchase additional credits in 115-generation increments for $15.
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.
Hugging Face presents a dynamic training method using adversarial data to improve model robustness. The technique automatically generates hard examples during training to strengthen performance.
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 details the infrastructure and techniques used to train BLOOM, a 176B-parameter multilingual language model. Technical article covering distributed optimization, memory management, and large-scale training challenges.
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 introduces BLOOM, the world's largest open multilingual language model. Trained on 46 languages, BLOOM matches proprietary state-of-the-art models in performance while ensuring open accessibility.
Introductory guide to sentiment analysis on Twitter using Hugging Face models. Covers loading datasets, fine-tuning pre-trained models, and deploying a sentiment analysis solution.