Refiner model crashes Python with 12 GB VRAM & 16 GB RAM while base model runs 2048×2048 fine

#8
by makisekurisu-jp - opened

Even with 12 GB VRAM and 16 GB RAM, running the q3km-quantized refiner model still causes a Python crash. However, the base model can handle 2048 × 2048 even with q5km quantization without any issues, so why does it fail when switching to the refiner model?

https://huggingface.co/QuantStack/HunyuanImage-2.1-GGUF

https://openart.ai/workflows/M1oiYEYRVhhXslYGSW7T

Sign up or log in to comment