The IQ4_XS_L quant is meant to be run with IK_Llama.cpp or the incoming EsoCroK.Cpp, because it uses a Q6_0 quant specific to IK_Llama.cpp
Here's EsoCrok, compatible with the llama.cpp mainline quants AND Q6_0
The usual Croco has not been updated (yet?) to support properly GLM 4.5 (or OpenAI GPT OSS)
