Prince-1

/

Medgemma-4b-pt-Onnx

like 0

I convert the models to onnx using onnxruntime-genai

pip install onnxruntime-genai --pre
pip install onnx onnx-ir safetensors torch transformers gguf

then run the following command

from onnxruntime_genai.models.builder import create _model

create_model("hf id name"
,"None /model dir/gguf)
,"output dir ",
," precision int4 f16 "
," cpu /cuda/webgpu",
,"cache_dir",
**extra)

https://github.com/microsoft/onnxruntime-genai/blob/main/src/python/py/models/README.md

I convert the models to onnx using onnxruntime-genai

pip install onnxruntime-genai --pre
pip install onnx onnx-ir safetensors torch transformers gguf

then run the following command

from onnxruntime_genai.models.builder import create _model

create_model("hf id name"
,"None /model dir/gguf)
,"output dir ",
," precision int4 f16 "
," cpu /cuda/webgpu",
,"cache_dir",
**extra)

Just change the inputs

Not all model types are supported

onnxruntime_genai\models\builder.py

Thanks for your reply! Since both MedGemma-4b-pt and MedGemma-4b-it use the Gemma3ForConditionalGeneration architecture, the builder will generate only the text component of the model.

Vision onnx model is still in development