#CO₂ Tracking & Offset Option
1 messages · Page 1 of 1 (latest)
Do you need help with something? You can contact support to get help if needed.
Most of the footprint is due to training so it s a bit hard to allocate it at inference time... besides mistral mostly uses nuclear power so it s more about compensating for putting future generations and civilizations at risk than being able to calm down temperatures by paying a few bucks...
If you re interested by the topic, you can search for "co2" on this discord as it s been discussed before.
Estimated last year that 1 request on mistral has comparable footprint as 1 organic soy bean... you could have a jar of beans next to your computer and every time you make a request you take one bean out. You ll have a good idea of where your footprint is.
I like this idea, it kind of forces to think about what you're doing, I can imagine that at the beginning you don't care, but then when you end up having to buy 18kg bags of soy beans you think twice before pressing enter 🙂
also, it might sound stupid, but do you know if it's the carbon footprint of cultivating one soy bean (and if so, does organic allow the use of trucs, in that case is that accounted for in the footprint of the bean)? or the amount of carbon one soy bean stores?
there's codecarbon, in theory it should be possible to have an estimate. or someone could run the model and a few prompts locally and estimate the carbon footprint per token
I'm not sure I understand the question. It s the footprint of buying organic soy beans at the store (cultivation and transport). I used le chat back then to do the calculation. You can find the full conversation by searching for "co2" on the discord.
#chat message
Not sure the numbers are still correct as models have gotten bigger and context size as well
thanks!
Here s more #ai-discussions message
What s clear is that these numbers are rising fast due to bigger models and bigger context windows and launching more and more autonomous tasks.
Openai makes great models I hear... but I don't have enough beans 🤣
Hey bud. If I am not mistaken, Mistral inference is done in Sweden (or is that training). Sweden uses hydro and wind, so 68% from renewables, 30% nuclear. Mistral is pretty darn clean. But yes, it would be nice if Mistral could estimate inference energy usage. All about being green here as well.
Yeah, I'm not sure what s done where... inference is a fraction of the footprint... but at the end of the day, the final order of magnitude is that American AI is 100x the footprint
And same for Chinese apparently
Yeah, American and Chinese AI pollute a lot more.
China is now 52% thats capacity not whats actually in the grid
In Germany it‘s around 55% (no nuclear)
there's gotta be a better name than "clean energy" for renewables+nuclear 🤣 low-carbon maybe?
Yeah, lol Nobody wants the nuclear waste
Wildlife apparently loves those Chernobyl and Fukushima sites, now that they are peaceful again
There are some intersting developments with some of the animals/plants there
lol I'm nowhere near chernobyl and I have dandelions growing like that
Oof. Do you live near a chemical factory?
lol nope 🙂
I didn't see many but found some in a field
Hmm. If you know any local biologists or universities, maybe they can take a look at those plants. Might be interesting for them to study
Oh that's a good idea, I'll think about it next time