3 ms·32K-token context length embedding models1 points by fzliu 2y agoskeptrune 2y agoI don't understand the use-case for these. We have always found it better to embed many smaller chunks and average the output as a kind of low-pass filter.