Why isn't there a UD-IQ4_XS?

#6
by neavo - opened

The UD-IQ4_K_XL is still significantly larger than IQ4XS, so why did the UD version skip quantizing IQ4XS?
The test results on the blog didn't compare UD-Q3_K_XL with IQ4_XS either. How should I choose?

Unsloth AI org

You can substitute the IQ4_XS one Q3_K_XL. As some layers in the Q3 one are larger so it evens out

You can substitute the IQ4_XS one Q3_K_XL. As some layers in the Q3 one are larger so it evens out

thanks for your reply

Sign up or log in to comment