Files are engineered to balance performance and memory footprints using high-precision F16 tensor swaps for sensitive layers (patch_embedding, dense_embedding, and fm_head) to preserve generation quality:

Q8_0 (~21.6 GB) — Near-lossless quality.
Q6_K (~17.5 GB) — High fidelity.
Q5_K_M (~15.5 GB) — Balanced tier.
Q4_K_M (~13.5 GB) — Standard compressed tier.
Q3_K_M (~10.5 GB) — Recommended tight VRAM profile for 8GB cards.
Q2_K (~9.4 GB) — Maximum compression.

Required Custom Node Fork

To run these GGUF files natively without structural AttributeError crashes or shape mismatches, use the dedicated node fork:

👉 GitHub: ComfyUI_SenseNova_U1_REBEL

Installation & Usage

Place your chosen .gguf file into your ComfyUI models directory (e.g., ComfyUI/models/gguf/).
Ensure you have installed the ComfyUI_SenseNova_U1_REBEL custom node suite.
Load the model using the SenseNova_SM_Model loader node and hook it up to the SenseNova_SM_Sampler.

realrebelai/SenseNova-U1.5-8B_GGUFs

SenseNova U1.5 8B MoT - GGUF Quantizations

Available Quants

Required Custom Node Fork

Installation & Usage

realrebelai/
SenseNova-U1.5-8B_GGUFs