Sort:  

Hello again — the official Qwopus-GLM5.1 frankenmerged 18B is live!

12–16GB GPU users, this one targets that hardware

The healing fine-tune fixed the seam between the two merged models beautifully

It was unusable before the healing run, but now it's surprisingly excellent, offering a neat blend of Qwopus and GLM 5.1 intelligence in a compact model

This leaned heavily on a collaborator's two Qwen 9b fine-tunes: Qwopus3.5-9B-v3.5 and Qwen3.5-9B-GLM5.1-Distill-v1

Outputs include a one-shot snake game and several frontend tests; raw results and a demo video are included in the repository

Full documentation of the merge and finetune-heal method is provided in the repository

All work was performed locally on an RTX 5090