Tag
Jackrong released Qwopus-GLM-18B-Merged-GGUF, a 64-layer frankenmerge combining two Qwen3.5-9B finetunes into an ~18B parameter model, healed with 1000-step LoRA fine-tuning to fix layer boundary issues. The model achieves 90.9% on capability benchmarks while using less than half the VRAM of Qwen 3.6-35B MoE.