saved from huihui's repo before the Q3_K_M was deleted
https://huggingface.co/huihui-ai/Huihui-GLM-4.5-Air-abliterated-GGUF/tree/main/Q3_K_M-GGUF
useful for 64gb ram
this is the setup I use with 8gb vram
koboldcpp.exe --host 0.0.0.0 --port 5001 --model Q3_K_M-GGUF-00001-of-00006.gguf --flashattention --contextsize 57344 --gpulayers 22 --moecpu
or
koboldcpp.exe --host 0.0.0.0 --port 5001 --model Q3_K_M-GGUF-00001-of-00006.gguf --flashattention --contextsize 16384 --gpulayers 48 --moecpu
this model is really good but i recommend mlabonne's abliterated gemma 3 27B for cases when GLM struggles with prompt adherence
Update
the original model has been deleted. here are the safetensors upconverted from Q4_K_M