amd/gemma-4-e4b-npu-eager

Gemma-4-E4B-IT ONNX export

Required Packages

• Mobius (main) - Install from source using pip install -e .
• Olive (main) - Install from source using pip install -e .
• Olive-recipe (main) - Run the recipe: https://github.com/microsoft/olive-recipes/tree/main/google-gemma-4-E2B-it
  • Change the input model name to gemma-4-E2B-it

Quantization

• K-quant
• Group size 32

cpu models:

• https://huggingface.co/amd/gemma-4-e4b-it-mobius-int4
• https://huggingface.co/amd/gemma-4-e2b-it-mobius-int4

model-generate command used :

"C:\ProgramData\miniforge3\envs\hybrid-llm\Scripts\model_generate.exe" --npu --eager --input D:\amd\gemma-4-e4b-it-mobius-int4 --output D:\amd\gemma-4-e4b-npu-eager --model-name decoder/model.onnx --extra_options model_type=gemma4