Using a PyTorch transformer for time series forecasting at inference time where you don’t know the decoder input — In this post, I will show how to run inference with a PyTorch transformer for time series forecasting. Specifically, we will use the PyTorch time series transformer I described in my previous post How to make a Transformer for time series forecasting with PyTorch. The post is structured as follows…

This post will show you how to transform a time series Transformer architecture diagram into PyTorch code step by step — Transformer models have shown state of the art performance in a number of time series forecasting problems [1][2][3]. In this post, you will learn how to code a transformer architecture for time series forecasting in PyTorch. Specifically, we’ll code the architecture used in the paper “Deep Transformer Models for Time…

Debates about bias in AI often overlook an important question: Should AI be used for a given decision in the first place? — Fortunately, bias in artificial intelligence (AI) is receiving a lot of attention these days. However, much of the debate that arises from stories about biased AI is about how to fix the data such that they are no longer biased, or whether inherently interpretable machine learning models should be used…

Here’s a curated list of 8 great podcast episodes about the environmental footprint of machine learning and how to reduce it. — Training and serving machine learning models can require a lot of electricity, and as long as not all electricity stems from renewable sources, the field of machine learning will be responsible for the emission of a significant amount of greenhouse gasses. For instance, it is estimated that the carbon footprint…

An estimate of ChatGPT’s costs supports estimate that ChatGPT uses millions of kilowatt hours per month. — I recently published an article in which I estimate ChatGPT’s electricity consumption in January 2023 to be between 1.1M and 23M KWh. 23M KWh is roughly the same amount of energy that 175,000 Danes use in a month on average. ChatGPT’s Electricity Consumption ChatGPT may have consumed as much electricity as 175,000 people in January 2023.towardsdatascience.com

Facebook used 2.6 million KWh hours of electricity and emitted 1,000 tons of CO2 when developing their new LLaMA models. — Facebook recently released four new language models called LLaMA, which outperform GPT-3 — ChatGPT’s underlying language model — on a number of tasks. Following a recent trend on resource consumption transparency in LLM training, Facebook estimate the electricity consumption and carbon footprint of their models in the paper published on…

ChatGPT may have consumed as much electricity as 175,000 people in January 2023. — I recently wrote an article in which I guesstimated ChatGPT’s daily carbon footprint to the around 24 kgCO2e. With little information available at the time about ChatGPT’s user base, my estimate was based on the assumption that the ChatGPT service was running on 16 A100 GPUs. …

This article attempts to estimate the carbon footprint of the popular OpenAI chatbot called ChatGPT — UPDATE March 20, 2023: In this blog post, I assumed that ChatGPT used 16 GPUs. Given ChatGPT’s popularity, this number has now been estimated to be upwards of 29,000 [10]. There’s a lot of talk about ChatGPT these days, and some people talk about the monetary costs of running the…

This article shows how to produce multi-step time series forecasts with XGBoost with 24h electricity price forecasting as an example. — A number of blog posts and Kaggle notebooks exist in which XGBoost is applied to time series data. However, it has been my experience that the existing material either apply XGBoost to time series classification or to 1-step ahead forecasting. This article shows how to apply XGBoost to multi-step ahead…