Multimodal model

Huihui 4 8B A4B v2

Huihui4-8B-A4B-v2 is a lightweight pruned MoE conversational model derived from Gemma-4-26B-A4B-it. It is designed for consumer-grade hardware and code-related conversation, using 32 experts with 8 active per token and adopting GLM-5.1-style thinking-mode data during supervised fine-tuning.

Type
Multimodal
Weights
Open
Downloads
82
Likes
20
Pipeline
image text to text
License
apache-2.0

Model hub

huihui-ai/Huihui4-8B-A4B-v2

Repo
huihui-ai/Huihui4-8B-A4B-v2
Library
transformers
Access
Public
Technical tags and languages
transformers safetensors gemma4 image-text-to-text sft Moe Pruned conversational dataset:huihui-ai/GLM-5.1-Multilingual-STEM base_model:google/gemma-4-26B-A4B-it base_model:finetune:google/gemma-4-26B-A4B-it endpoints_compatible
Open on Hugging Face

More from HuiHui ai

10