File size: 677 Bytes
a11dbf4 07b2b3b a11dbf4 c4ef27c a11dbf4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 | ---
license: creativeml-openrail-m
base_model:
- NucleusAI/Nucleus-Image
tags:
- text-to-image
- fp8
---
# Nucleus-Image-FP8-Native
This is a native FP8 (float8_e4m3fn) quantization of the 17B Nucleus-Image model.
## ⚠️ VRAM Warning
This model is **EXTREMELY HEAVY**. Even in FP8, the weights alone take up ~13-14GB of VRAM.
- **16GB VRAM (RTX 4060 Ti / 4070 Ti Super / 4080):** Recommended to use with `sequential_cpu_offload` for stability. Pure GPU inference might OOM at 1024x1024.
- **24GB VRAM (RTX 3090 / 4090):** Best experience. Can run Pure GPU without offloading.
- **12GB VRAM or less:** **NOT RECOMMENDED** unless using heavy CPU offloading (will be slow). |