IDK if I'll publish the crate on crates.io. I plan to use some nightly features to achieve what I want, and I'll likely implement it directly in my current project. However, I wanted to share some of my work with those who might be interested in this topic.

You're welcome!

There are likely more resources on this topic, but it might still be a bit early to have a complete overview of the current situation, unfortunately...

The article is interesting but likely overlooks or omits some key points. Also, since it’s a Meta paper, I’d suggest taking it with a grain of salt.

These two articles are also worth a look:
- mit-genai.pubpub.org/pub/8ulgrckc...
- environment.yale.edu/news/article...

Companies running AI models focus on energy sources ("we use green energy, we'll use nuclear") but ignore lifecycle analysis. The biggest impact likely comes from semiconductors, water usage, mining etc.

I found a Meta paper talking about the AI carbon footprint:
arxiv.org/pdf/2111.00364