├── .gitignore ├── Cloze.ipynb ├── R ├── README.md ├── data │ ├── data_for_sphere_plots │ │ ├── 13b_zh_lines.csv │ │ ├── 70b_zh_lines.csv │ │ └── 7b_zh_lines.csv │ ├── inputs │ │ ├── 13b_de_de_dataset.csv │ │ ├── 13b_fr_fr_dataset.csv │ │ ├── 13b_ru_ru_dataset.csv │ │ ├── 70b_de_de_dataset.csv │ │ ├── 70b_de_zh_dataset.csv │ │ ├── 70b_fr_fr_dataset.csv │ │ ├── 70b_ru_ru_dataset.csv │ │ ├── 7b_de_de_dataset.csv │ │ ├── 7b_de_zh_dataset.csv │ │ ├── 7b_fr_fr_dataset.csv │ │ ├── 7b_ru_ru_dataset.csv │ │ ├── clean │ │ │ ├── de_de.csv │ │ │ ├── de_zh.csv │ │ │ ├── fr_fr.csv │ │ │ └── ru_ru.csv │ │ └── unique │ │ │ ├── de_de.csv │ │ │ ├── fr_fr.csv │ │ │ └── ru_ru.csv │ ├── latents │ │ ├── .gitignore │ │ ├── probe_input.csv │ │ ├── probe_output.csv │ │ ├── rmsnorm_13b.csv │ │ ├── rmsnorm_70b.csv │ │ └── rmsnorm_7b.csv │ ├── logit_ranks │ │ ├── ranks_13b.tsv │ │ ├── ranks_70b.tsv │ │ ├── ranks_7b.tsv │ │ ├── relranks_13b.tsv │ │ ├── relranks_70b.tsv │ │ └── relranks_7b.tsv │ ├── tokens.json │ └── tokens.tsv ├── embed_unembed.r └── plots │ ├── convert_script.txt │ ├── entropy_70b.pdf │ ├── entropy_70b.png │ ├── lang_probs_70b.pdf │ ├── lang_probs_70b.png │ ├── mds.pdf │ ├── mds.png │ ├── token_energy_70b.pdf │ └── token_energy_70b.png ├── README.md ├── Translation.ipynb ├── data └── langs │ ├── de │ ├── clean.csv │ └── text.csv │ ├── en │ ├── clean.csv │ └── text.csv │ ├── fr │ ├── clean.csv │ └── text.csv │ ├── ru │ ├── clean.csv │ └── text.csv │ └── zh │ └── clean.csv ├── llamawrapper.py ├── plot_grid.sh ├── requirements.txt └── utils.py /.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/.gitignore -------------------------------------------------------------------------------- /Cloze.ipynb: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/Cloze.ipynb -------------------------------------------------------------------------------- /R/README.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/README.md -------------------------------------------------------------------------------- /R/data/data_for_sphere_plots/13b_zh_lines.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/data_for_sphere_plots/13b_zh_lines.csv -------------------------------------------------------------------------------- /R/data/data_for_sphere_plots/70b_zh_lines.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/data_for_sphere_plots/70b_zh_lines.csv -------------------------------------------------------------------------------- /R/data/data_for_sphere_plots/7b_zh_lines.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/data_for_sphere_plots/7b_zh_lines.csv -------------------------------------------------------------------------------- /R/data/inputs/13b_de_de_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/13b_de_de_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/13b_fr_fr_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/13b_fr_fr_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/13b_ru_ru_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/13b_ru_ru_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/70b_de_de_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/70b_de_de_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/70b_de_zh_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/70b_de_zh_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/70b_fr_fr_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/70b_fr_fr_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/70b_ru_ru_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/70b_ru_ru_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/7b_de_de_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/7b_de_de_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/7b_de_zh_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/7b_de_zh_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/7b_fr_fr_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/7b_fr_fr_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/7b_ru_ru_dataset.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/7b_ru_ru_dataset.csv -------------------------------------------------------------------------------- /R/data/inputs/clean/de_de.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/clean/de_de.csv -------------------------------------------------------------------------------- /R/data/inputs/clean/de_zh.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/clean/de_zh.csv -------------------------------------------------------------------------------- /R/data/inputs/clean/fr_fr.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/clean/fr_fr.csv -------------------------------------------------------------------------------- /R/data/inputs/clean/ru_ru.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/clean/ru_ru.csv -------------------------------------------------------------------------------- /R/data/inputs/unique/de_de.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/unique/de_de.csv -------------------------------------------------------------------------------- /R/data/inputs/unique/fr_fr.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/unique/fr_fr.csv -------------------------------------------------------------------------------- /R/data/inputs/unique/ru_ru.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/inputs/unique/ru_ru.csv -------------------------------------------------------------------------------- /R/data/latents/.gitignore: -------------------------------------------------------------------------------- 1 | *.pt 2 | 3 | -------------------------------------------------------------------------------- /R/data/latents/probe_input.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/latents/probe_input.csv -------------------------------------------------------------------------------- /R/data/latents/probe_output.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/latents/probe_output.csv -------------------------------------------------------------------------------- /R/data/latents/rmsnorm_13b.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/latents/rmsnorm_13b.csv -------------------------------------------------------------------------------- /R/data/latents/rmsnorm_70b.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/latents/rmsnorm_70b.csv -------------------------------------------------------------------------------- /R/data/latents/rmsnorm_7b.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/latents/rmsnorm_7b.csv -------------------------------------------------------------------------------- /R/data/logit_ranks/ranks_13b.tsv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/logit_ranks/ranks_13b.tsv -------------------------------------------------------------------------------- /R/data/logit_ranks/ranks_70b.tsv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/logit_ranks/ranks_70b.tsv -------------------------------------------------------------------------------- /R/data/logit_ranks/ranks_7b.tsv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/logit_ranks/ranks_7b.tsv -------------------------------------------------------------------------------- /R/data/logit_ranks/relranks_13b.tsv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/logit_ranks/relranks_13b.tsv -------------------------------------------------------------------------------- /R/data/logit_ranks/relranks_70b.tsv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/logit_ranks/relranks_70b.tsv -------------------------------------------------------------------------------- /R/data/logit_ranks/relranks_7b.tsv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/logit_ranks/relranks_7b.tsv -------------------------------------------------------------------------------- /R/data/tokens.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/tokens.json -------------------------------------------------------------------------------- /R/data/tokens.tsv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/data/tokens.tsv -------------------------------------------------------------------------------- /R/embed_unembed.r: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/embed_unembed.r -------------------------------------------------------------------------------- /R/plots/convert_script.txt: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/convert_script.txt -------------------------------------------------------------------------------- /R/plots/entropy_70b.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/entropy_70b.pdf -------------------------------------------------------------------------------- /R/plots/entropy_70b.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/entropy_70b.png -------------------------------------------------------------------------------- /R/plots/lang_probs_70b.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/lang_probs_70b.pdf -------------------------------------------------------------------------------- /R/plots/lang_probs_70b.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/lang_probs_70b.png -------------------------------------------------------------------------------- /R/plots/mds.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/mds.pdf -------------------------------------------------------------------------------- /R/plots/mds.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/mds.png -------------------------------------------------------------------------------- /R/plots/token_energy_70b.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/token_energy_70b.pdf -------------------------------------------------------------------------------- /R/plots/token_energy_70b.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/R/plots/token_energy_70b.png -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/README.md -------------------------------------------------------------------------------- /Translation.ipynb: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/Translation.ipynb -------------------------------------------------------------------------------- /data/langs/de/clean.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/de/clean.csv -------------------------------------------------------------------------------- /data/langs/de/text.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/de/text.csv -------------------------------------------------------------------------------- /data/langs/en/clean.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/en/clean.csv -------------------------------------------------------------------------------- /data/langs/en/text.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/en/text.csv -------------------------------------------------------------------------------- /data/langs/fr/clean.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/fr/clean.csv -------------------------------------------------------------------------------- /data/langs/fr/text.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/fr/text.csv -------------------------------------------------------------------------------- /data/langs/ru/clean.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/ru/clean.csv -------------------------------------------------------------------------------- /data/langs/ru/text.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/ru/text.csv -------------------------------------------------------------------------------- /data/langs/zh/clean.csv: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/data/langs/zh/clean.csv -------------------------------------------------------------------------------- /llamawrapper.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/llamawrapper.py -------------------------------------------------------------------------------- /plot_grid.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/plot_grid.sh -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/requirements.txt -------------------------------------------------------------------------------- /utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/epfl-dlab/llm-latent-language/HEAD/utils.py --------------------------------------------------------------------------------