PyTorch’s Post

See below for an opportunity to leverage the PyTorch torch.compile feature to achieve up to 2x better performance for Hugging Face model inference and up to 1.35x better performance for TorchBench model inference compared to the previous PyTorch eager mode. See below for a detailed blog - happy inferencing! #AWS #Graviton #PyTorch #AIML

Performance by default on #Neoverse with improvements to the Open Source PyTorch implementation from AWS. torch.compile is a simple interface to unlock more performance for machine learning on Arm.

Amazon Web Services (AWS) has optimized the PyTorch torch.compile feature for AWS Graviton3 processors. Congratulations, Sunita, on this impressive work! Stay tuned for more details on leveraging these optimizations for faster ML workflows! #MachineLearning #PyTorch #AWS #Graviton3

Deep dive: PyTorch 2.0 on AWS Graviton. Watch this AWS Online TechTalk video which will talk through the price performance benefits of your PyTorch 2.0 workloads on AWS. AWS Graviton based instance inference performance for PyTorch 2.0 is up to 3.5x faster than the previous PyTorch release.

It was a crazy lesson today, I learned how to set up Auto Scaling Groups in AWS. It's fascinating how you can use Machine Learning to predict your future traffic and have it scale your instances automatically. #1dayatatime

Why you should care about fine-tuning smaller models: they are better, cheaper, and faster than large behemoths. A finetuned SQLCoder-7B costs you just $300/mo to host on AWS, and handles up to 1600 requests/hour on a single machine At the same scale, GPT-4 costs 60x more at $18,000/mo

Check out my most recent article on Medium, A Guide to Passing the AWS ML Exam! I shared my experience with the exam as well as tips to help you pass it too. #awscertified #aws #machinelearning

Inferentia 2 accelerators reduce the cost of inferencing for sophisticated Deep Learning models. Having them available in Europe regions will reduce the cost of many projects I have done so far.