Im always so surprised that embedding models we had for years like minlm (80mb) are so small, and I really wonder why not more on device searches use something

Im always so surprised that embedding models we had for years like minlm (80mb) are so small, and I really wonder why not more on device searches use something like it.