nntsuzu/Kikka-multimodal-2B

Model Card for Model ID

It is new multimodal model that merged Gemma4 and Lfm2.5 architechtures.

Model Details

Model Description

• Developed by: nn-tsuzu
• Model type: multimodal
• Languages: English, German, Italian, Franch, Spanish, Chinese, Japanese, Korean
• Supported inputs: Text, Image, Video, Audio

Merged model archtechtures

• Parts of text: LiquidAI/LFM2.5-1.2B-Instruct
• Parts of vision: google/gemma-4-E2B-it
• Parts of audio: google/gemma-4-E2B-it

How to use

Please install torch, torchcodec, transformers>=5.5.0, and libsora.

conda install "ffmpeg"
# or
conda install "ffmpeg" -c conda-forge

pip install torch torchcodec

pip install transformers libsora