10Eros-Max-Turbo-Beta7 (GGUF Quantizations)

This repository provides community-quantized GGUF weights for 10Eros-Max-Turbo-Beta7, an experimental 8-step turbo hybrid model created by TenStrip based on the MiniMax-H3 architecture.


πŸš€ What's New in Beta 7?

Beta 7 refines the merge methodology by converting LoRA data into MLP-only weights merged with a magnitude cap, leaving the base MiniMax-H3 attention layers and reference gates 100% intact. This enables ultra-fast 8-step turbo sampling while preserving deep texture fidelity, realistic fluid physics, and native audio-visual synchronization.


πŸ“¦ Available GGUF Quantizations

Quantization File Size Quality / Notes
Q3_K_M 8.9 GB Balanced ultra-low VRAM option.
Q4_K_S 11.6 GB Smaller footprint Q4 variant.
Q4_K_M 11.6 GB 🌟 The Community Golden Standard (Best balance of speed, VRAM, and micro-texture).
Q5_K_S 14.1 GB High-fidelity 5-bit lower tier.
Q5_K_M 14.1 GB Near-lossless visual texture retention.

πŸ› οΈ ComfyUI Integration Guide

To run these GGUF files in ComfyUI:

  1. Ensure you have ComfyUI-GGUF installed (git clone https://github.com/city96/ComfyUI-GGUF in your custom_nodes directory).
  2. Download your preferred .gguf file (e.g., 10Eros-Max-Turbo-Beta7-Q4_K_M.gguf) and place it in your ComfyUI/models/diffusion_models/ folder.
  3. Load the model using the UnetLoaderGGUF node.

Recommended Generation Settings (Beta 7 8-Step Turbo)

  • Resolution: 1280 x 720 (720p) or 1920 x 1088 (1080p native sweet spot).
  • Steps: Exactly 6 to 8 steps (Turbo optimized).
  • CFG Scale: 1.0 to 1.5 (Higher CFG values will oversaturate or burn the output).
  • Sampler / Scheduler: res_multistep or euler with simple scheduler.
  • Important: Keep SLA sparsity, Spectrum, and TeaCache disabled (they interfere with H3 hybrid gate accuracy).

βš–οΈ Licensing & Attribution

  • Base Architecture: MiniMax H3, licensed under the MiniMax H3 Community License Agreement (Copyright Β© 2026 MiniMax).
  • Model Weights: 10Eros-Max by TenStrip.
  • GGUF Conversion & Packaging: Maintained by Abiray.
Downloads last month
9,287
GGUF
Model size
20B params
Architecture
wan
Hardware compatibility
Log In to add your hardware

3-bit

4-bit

5-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for Abiray/10Eros-Max-Turbo-Beta7-GGUF

Quantized
(9)
this model