MiniMax-H3 10Eros-Max beta6 pruned (8-step TURBO) — W4A8
W4A8 quantization of 10Eros_Max_h3_8stepTURBO_beta6_pruned (bf16 source: TenStrip/10Eros-Max, 10Eros_Max_h3_8stepTURBO_beta6_pruned.safetensors) — the 8-step turbo build (TenStrip/10Eros-Max).
- Format:
asym_w4a8_int8,group_size=16,convrot_groupsize=256, per-tensor Lloyd-Max codebook + fp8 group scales (Kijai / comfy-kitchen W4A8 layout). - Size: ~11.68 GB — 68.8% smaller than the ~37.5 GB bf16 source; fits a 16 GB card with headroom.
- Layers: 200 2D-linear layers quantized (96% of policy-targeted bytes); norms / first / last kept higher-precision (332 kept).
- Quantizer: comfyui-mixed-quantizer
--format w4a8 --group-size 16 --codebook-mode fit; the quantizer's full validation passes (421 passed, 0 failed). - Verified: loads via UNETLoader (convrot) and generates coherent video with speech in ComfyUI.
- sha256:
96c5d85b1f658d820e2a66e33f41baf31002e7f838eaa2bf6487bd43fc94bbf8
Requirements
- ComfyUI ≥ v0.31.0 (native W4A8 loader) or the
comfyui_w4a8_loader.patch. - comfy-kitchen with
AsymW4A8Int8Layout(PR #90). - CUDA SM ≥ 8.0.
Inherits the source model's license.