opening-up-chatgpt / opening-up-chatgpt.github.io

Tracking instruction-tuned LLM openness. Paper: Liesenfeld, Andreas, Alianda Lopez, and Mark Dingemanse. 2023. “Opening up ChatGPT: Tracking Openness, Transparency, and Accountability in Instruction-Tuned Text Generators.” In Proceedings of the 5th International Conference on Conversational User Interfaces. doi:10.1145/3571884.3604316.
https://opening-up-chatgpt.github.io/
Apache License 2.0
80 stars 5 forks source link

Add training time & hardware reported (towards energy consumption measures) #55

Open mdingemanse opened 9 months ago

mdingemanse commented 9 months ago

Djoerd Hiemstra notes (on Mastodon):

als je opening up chatgpt uitbreidt met een kolom "training time reported" en "hardware reported" (en de code is openbaar) dan kunnen we een schatting doen van de hoeveelheid energy dat het (eenmaal) trainen van het model heeft gekost, plus de hoeveelheid CO2 die daarbij uitgestoten werd.

This would be an interesting experiment, perhaps best done in a fork or a separate branch to see (i) how much data there is and (ii) how much two additional columns mess up the website layout.

Would require adding two more groups of key/value pairs in the yaml template (and in all existing YAML files) and adding those same fields (and default weights) to consolidate_csv.py.

training_reported:
    class: closed
    link:
    notes:

hardware_reported:
    class: closed
    link:
    notes: