In this video, I compare the cost-performance of AWS Trainium, a new custom chip designed by AWS, with NVIDIA A10G GPUs. I first launch a trn1.32xlarge instance (16 Trainium chips) and a g5.48xlarge (8 A10Gs). Then, I run a natural language processing job, fine-tuning the BERT Large model on the full Yelp review datatset. I use the BF16 data format with the maximum sequence length supported by the model (512). The results? The Trainium job is 5x faster. As the trn1 instance is only 30% more expensive, this is a huge improvement in cost-performance!

In this video, I show you how to accelerate Transformer training with AWS Trainium, a new custom chip designed by AWS, and the brand new Hugging Face Neuron Amazon Machine Image (AMI). First, I walk you through the setup of an Amazon EC2 trn1.32xlarge instance, equipped with 16 Trainium chips. Then, I run a natural language processing job, accelerating a BERT model to classify the Yelp review datatset on 32 Neuron cores.

Witty Works builds a digital writing assistant for inclusive language. We discussed text classification, few-shot learning, the benefits of the SetFit library, and how Hugging Face helped get to production quicker :)

Hyperproductive ML with Hugging Face in action! With just a few lines of code, here’s how I trained a highly accurate language identification model, and then optimized its latency 3x with Optimum Intel and Openvino 😎

Not one, but three new Hugging Face videos today :) Learn and share! In this video, you will learn how to accelerate PyTorch inference with an Intel Sapphire Rapids server running on AWS. Working with popular Hugging Face transformers implemented with PyTorch, we’ll first measure their performance on an…

Tabular data is everywhere, from spreadsheets to tables embedded in text documents. In this Hugging Face Space, you can use the Google TAPAS and Microsoft TAPEX models to ask questions in natural language on CSV data. Give it a try and let me know what you think :) ✅ Space…

The Hugging Face hub hosts 1,700+ image classification models. I’ve picked the top ones from Google, Meta, Microsoft, NVIDIA and Humphrey Shi. All of them have been fine-funed on the ImageNet-1k dataset, which makes it easy to compare their performance on your images. Enjoy! Space: https://huggingface.co/spaces/juliensimon/battle_of_image_classifiers

Multi-modal transformers are rising fast. A great example is the Audio Spectrogram Transformer, an audio classification model that was just added to the Hugging Face Transformers library. This model first creates a spectrogram image of an audio clip and then classifies the image with a Vision Transformer model. Amazing results! ✅ Spaces demo: https://huggingface.co/spaces/juliensimon/keyword-spotting ✅ Model: https://huggingface.co/MIT/ast-finetuned-speech-commands-v2 ✅ Paper: https://arxiv.org/abs/2104.01778

What if you could ask questions on HTML documents, without having to convert them to plain text first? Well, that’s exactly the purpose of the Microsoft MarkupLM: just grab a page and ask a question. I’ve built a Hugging Face Space to let you experiment with any live URL. I also implemented multithreading to speed things up on CPU. Give it a go and let me know what you think :) ✅ Space: https://huggingface.co/spaces/juliensimon/webpage_questions ✅ Model: https://huggingface.co/microsoft/markuplm-base-finetuned-websrc ✅ Paper: https://arxiv.org/abs/2110.08518