Q4_K_XL has smaller size than Q4_K_M.

#10
by lzm1066258 - opened

image
Why is the size of Q4_K_XL smaller than that of Q4_K_M? Is there also a quantification problem similar to that of the earlier quant version of qwen3.5?

I'm also curious about this question.πŸ€”

AFAIK: UD=Unsloth Dynamic, and "XL" is also Unsloth's own recipe, so I think that's why.
About the differences, maybe there's some info in Unsloth's page?

Sign up or log in to comment