MXFP4?
#1
by floory - opened
i know gpt-oss-20b should not be used with other quantisations other than its original due to performance loss. i saw https://huggingface.co/mradermacher/rsi-gpt-oss-20b-i1-GGUF has one, but this updated version does not.
additional question: assuming you upload MXFP4 versions of both static and imatrix quants - would it be worse to use imatrix quants since it wouldn't be "as original", or is it still better? if you have absolutely no clue i don't blame you! :D