AI
Hugging Face Reads, Feb. 2021 - Long-range Transformers
Hugging Face's blog post discusses the limitations of traditional Transformer models in processing long-range dependencies and proposes a new architec...
AI
AI
Multimodal neurons in artificial neural networks
Researchers have found a type of neuron in the artificial neural network CLIP that responds to the same concept regardless of how it's presented. This...
AI
Simple considerations for simple people building fancy neural networks
The article discusses best practices for building and training neural networks, focusing on simplicity and ease of use. It highlights the importance o...
AI
Retrieval Augmented Generation with Huggingface Transformers and Ray
Hugging Face, a popular AI model hub, has released a new tutorial on using their transformers in conjunction with the distributed computing framework ...
AI
Hugging Face on PyTorch / XLA TPUs
Hugging Face, a popular AI model hub, has announced that its models can now be run on Google's PyTorch/XLA platform for TPUs. This allows users to dep...
AI
Understanding the capabilities, limitations, and societal impact of large language models
Large language models have become increasingly sophisticated, but their capabilities and limitations are not yet fully understood. They can generate h...
AI
Faster TensorFlow models in Hugging Face Transformers
Hugging Face has released an update to its Transformers library, allowing for faster deployment of TensorFlow models. The change uses a new serving me...
AI
Scaling Kubernetes to 7,500 nodes
OpenAI has scaled Kubernetes clusters to 7,500 nodes. This infrastructure supports large models like GPT-3, CLIP, and DALL·E, as well as rapid small-s...
AI
Fit More and Train Faster With ZeRO via DeepSpeed and FairScale
Researchers have developed a new technique called ZeRO, which uses DeepSpeed and FairScale to improve the efficiency of training large AI models. ZeRO...
AI
How we sped up transformer inference 100x for 🤗 API customers
Hugging Face, a popular AI model repository and developer of the Transformers library, claims to have accelerated inference speeds for its 🤗 API custo...
AI
CLIP: Connecting text and images
OpenAI has introduced a neural network called CLIP that can learn visual concepts from natural language supervision. This means it can understand what...
AI
DALL·E: Creating images from text
OpenAI has developed a neural network called DALL·E, which can generate images based on text descriptions. The system is trained to understand a wide ...