---
base_model: Gryphe/Gemma-4-26B-A4B-StyleTune-V2
---

[EXL3](https://github.com/turboderp-org/exllamav3) quantization of [Gemma-4-26B-A4B-StyleTune-V2](https://huggingface.co/Gryphe/Gemma-4-26B-A4B-StyleTune-V2), 4 bits per weight.

Use [CPU offloading](https://github.com/turboderp-org/exllamav3/blob/master/doc/env_vars.md#cpu-moe-offload) by setting [`EXL3_MOE_CPU_OFFLOAD`](https://github.com/turboderp-org/exllamav3/pull/257).

E.g. with [TabbyAPI](https://github.com/theroyallab/tabbyAPI):

```
EXL3_MOE_CPU_OFFLOAD=10 python main.py ...
```