ubergarm

/

GLM-4.7-Flash-GGUF

like 22

Hello,

Thanks for these quants, I tried mxfp4 as it had the least perplexity and it seems to be working much better for me than the UD_Q5 from unsloth.

Command I ran:

      ${ik_llama_cpp}
      -m ${models_dir}/LLMs/GLM-4.7-Flash-MXFP4.gguf
      -a 'GLM_4.7_Flash'
      -ger --special
      --merge-qkv 
      -mla 3 -amb 512 
      -ngl 99 
      -c 100000
      --temp 0.7
      --top-p 1.0
      --min-p 0.01
      --jinja