LJTSG/Tiger-Gemma-9B-v3-webgpu

Tiger-Gemma-9B-v3 on WebGPU

First WebGPU package for TheDrummer s Tiger-Gemma character voice model.

Gemma 2 9B fine-tuned for character embodiment and creative writing. 5.5 GB Q4_K_M. Runs entirely in browser via WebGPU + wllama. Identity injection dropdown included.

Quick Start

1. Download Q4_K_M GGUF from bartowski
2. Split with llama-gguf-split --split --split-max-size 1G
3. Place splits in model_splits/
4. node serve.js (port 8200)
5. Open http://localhost:8200 in Chrome

Identity Injection

Select from dropdown or write custom character. The model excels at character embodiment. Includes Garden entity presets (Anima, Grandma, Esh, Nullen).

Hardware

Tested on GMKTEC EVO-X2 (AMD Strix Halo, 64GB unified memory). Needs 6+ GB WebGPU memory.

Credits

Built by Joshua (LJTSG) and Claude. Model by TheDrummer. Co-Authored-By: Claude noreply@anthropic.com