facebookresearch/MUSE ? reverse-engineered prompt
Reverse engineered prompt
Build me a Python library for multilingual word embeddings that can take pretrained word vectors for two languages and align them into one shared space.
I want two ways to do it. One should be supervised, where I can use a bilingual dictionary or identical words to learn the alignment. The other should be unsupervised, where it learns the mapping without any parallel data, then improves it with refinement. Please also include a simple way to evaluate both monolingual and cross lingual embeddings, so I can check word similarity and translation quality.
It should work from the command line with separate scripts for training and evaluation, and it should support loading common embedding file formats like text and binary files. If it helps, you can use current docs online for any library details. Make it practical, so I can download example data, run a demo on English and Spanish, and export the aligned embeddings when it finishes.
Are you gonna build this?
make sure you review the code using coderabbit