Working GPTQ-Int4 version?

#15
by koesn - opened

Still no official GPTQ-Int4 like Qwen/Qwen3-30B-A3B-GPTQ-Int4. This GPTQ-Int4 runs out of the box on LMDeploy backend. Unfortunately there's no working GPTQ-Int4 out there.

btbtyler09/Qwen3-30B-A3B-Instruct-2507-gptq-4bit version is not working, because it's said "group_size=32" which is not supported. Only "group_size=128" supported. Original Qwen/Qwen3-30B-A3B-GPTQ-Int4 is also 128.

Very much thank's and appreciation if someone add this GPTQ to Instruct-2507 and Thinking-2507.

Sign up or log in to comment