Some minor weird things I noticed

epfl-dlab / llm-latent-language

Repo accompanying our paper "Do Llamas Work in English? On the Latent Language of Multilingual Transformers".

51 stars 12 forks source link

Some minor weird things I noticed #1

Closed Butanium closed 4 months ago

Butanium commented 6 months ago

The output format of this one seems weird: https://github.com/epfl-dlab/llm-latent-language/blob/1be3cd6eaf14c7408e8f03703d1c1905a6b00c44/Translation.ipynb#L242 E.g. some column are called "de" but contains french words.
This line should use lang_latent and not 'en': https://github.com/epfl-dlab/llm-latent-language/blob/1be3cd6eaf14c7408e8f03703d1c1905a6b00c44/Translation.ipynb#L316

wendlerc commented 5 months ago

Right! Can you make a pull request?

The df_en_de_fr and its column names not matching the actual languages stem from the fact that we started with de and fr and then since the code is the same for other languages just replace the 'de' with the source language and 'fr' with the target language (or the other way around, not sure anymore).