Downloads · 30 days
0
MiniMax-H3-experimental
Jackxuanxuan/MiniMax-H3-experimental
MiniMax-H3-experimental is a machine learning model from Jackxuanxuan. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
w4a8 is experimental format using 4bit weights and int8-convrot activations, at testing phase, further info here:
Access
Public
Updated Sep 8, 2026
Repo size
66.4 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors66.4 GB · 100%
At a glance
- Access
- Public
- Created
- Sep 8, 2026
- Updated
- Sep 8, 2026
- SHA
- 831d1e48
Model card
From the Hugging Face model README
w4a8 is experimental format using 4bit weights and int8-convrot activations, at testing phase, further info here:
https://github.com/Comfy-Org/comfy-kitchen/pull/90
Requires ComfyUI 0.31.0
int8_convrot VAE needs ComfyUI 0.31.0 or you get black outputs
It speeds up VAE decode times by ~1.5x
ref lora is the difference between fl2va and ref2va, completely experimental, I don't even know if it has a use case at this point.
minimax_h3_fastvideo_vsa_datafree_1300step_4step_int8_convrot.safetensors
Currently for testing with these PRs:
- Created
- Sep 8, 2026
- Updated
- Sep 8, 2026