optimized word embedding sharing

[damon-demon]

Dear authors,

I am currently conducting research that aligns closely with the work presented in your ICLR'24 paper, specifically regarding the optimized word embeddings you developed. To ensure the accuracy of my analysis and to use these embeddings as a baseline, I would greatly appreciate it if you could share the checkpoints with me. This would help mitigate any discrepancies due to randomness or differences in code execution across various devices.

Thank you for considering my request. I look forward to your positive response. ☀️