Gemma 4 E4B IQ2_XXS quantization gains 140.54% reasoning

Tensor-level quantization allocation lifts Gemma 4 E4B reasoning scores from 28.9 to 69.5 at the same 3.3GB budget. The quantized model is available on HuggingFace as ByteOtter/gemma-4-E4B-it-CADA-IQ2_XXS.