| base_model: google/gemma-3-4b-it | |
| [EXL3](https://github.com/turboderp-org/exllamav3) quantization of [gemma-3-4b-it](https://huggingface.co/unsloth/gemma-3-4b-it), 8 bits per weight, including output layers. | |
| ### HumanEval (argmax) | |
| | Model | Q4 | Q6 | Q8 | FP16 | | |
| | -------------------------------------------------------------------------------------- | ---- | ---- | ---- | ---- | | |
| | [gemma-3-4b-it-exl3-8bpw-h8](https://huggingface.co/isogen/gemma-3-4b-it-exl3-8bpw-h8) | 72.0 | 73.2 | 71.3 | 70.1 | | |