@outsource_: NEW GLM+ QWEN 18B RUNS ON CONSUMER GPU IT BEATS 35B MoE AT HALF THE VRAM @KyleHessling1 just dropped the healed Qwopus-…
Summary
A new 18B merged quantized model, Qwopus-GLM-18B-GGUF, outperforms 35B MoE models while using half the VRAM and running on consumer GPUs.
View Cached Full Text
Cached at: 04/21/26, 10:32 AM
NEW GLM+ QWEN 18B RUNS ON CONSUMER GPU IT BEATS 35B MoE AT HALF THE VRAM @KyleHessling1 just dropped the healed Qwopus-GLM-18B-Merged-GGUF Insane 64-layer frankenmerge of two elite Qwen3.5-9B finetunes (Opus reasoning + GLM-5.1 distill). This thing is cooking on consumer
Similar Articles
24+ tok/s from ~30B MoE models on an old GTX 1080 (8 GB VRAM, 128k context)
A developer demonstrates running MoE models like Qwen 3.6 35B-A3B and Gemma 4 26B-A4B at 24+ tok/s on an old GTX 1080 (8GB VRAM) with 128k context using llama.cpp with MoE offloading and TurboQuant KV cache quantization, revealing optimization tricks for Gemma's MTP speculative decoding.
Running Qwen 3.8 next on 16vram+32ram - A useful/fun post for the gpu poors
A Reddit user shares how they successfully ran the Qwen 3.8 Next MoE model on a system with 16GB VRAM and 32GB RAM using aggressive quantization and specific llama.cpp settings, achieving usable performance for large models on limited hardware.
KyleHessling1/Qwopus-GLM-18B-Merged-GGUF
An experimental 18B-parameter model created by stacking two Qwen-3.5-9B finetunes and healing the layer boundary with 1000-step QLoRA; the resulting GGUF beats Qwen 3.6-35B MoE on a 44-test suite while fitting in 9.2 GB VRAM.
Jackrong/Qwopus-GLM-18B-Merged-GGUF
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.
@no_stp_on_snek: This is huge. 3090 class gpus rejoice!
Unsloth AI releases quantized Qwen3.6 models that run 2.5× faster on consumer GPUs, with the 27B model fitting in 24GB VRAM and the 35B-A3B achieving high throughput.