Working environment for running MedGemma (PyTorch/Transformers versions, CUDA, etc.)

HAI-DEF
,
1

Hi everyone,

I’m trying to get MedGemma running reliably and I’m running into version pinning issues (mainly around torch, transformers, and accelerator). Could folks who have MedGemma working end-to-end share their exact environment? Thanks!

1 Like
3

MedGemma should work with the current latest versions of transformers, torch, and accelerate:

accelerate==1.11.0
certifi==2025.11.12
charset-normalizer==3.4.4
filelock==3.20.0
fsspec==2025.10.0
hf-xet==1.2.0
huggingface-hub==0.36.0
idna==3.11
Jinja2==3.1.6
MarkupSafe==3.0.3
mpmath==1.3.0
networkx==3.5
numpy==2.3.4
nvidia-cublas-cu12==12.8.4.1
nvidia-cuda-cupti-cu12==12.8.90
nvidia-cuda-nvrtc-cu12==12.8.93
nvidia-cuda-runtime-cu12==12.8.90
nvidia-cudnn-cu12==9.10.2.21
nvidia-cufft-cu12==11.3.3.83
nvidia-cufile-cu12==1.13.1.3
nvidia-curand-cu12==10.3.9.90
nvidia-cusolver-cu12==11.7.3.90
nvidia-cusparse-cu12==12.5.8.93
nvidia-cusparselt-cu12==0.7.1
nvidia-nccl-cu12==2.27.5
nvidia-nvjitlink-cu12==12.8.93
nvidia-nvshmem-cu12==3.3.20
nvidia-nvtx-cu12==12.8.90
packaging==25.0
pillow==12.0.0
pip==25.3
psutil==7.1.3
PyYAML==6.0.3
regex==2025.11.3
requests==2.32.5
safetensors==0.6.2
setuptools==80.9.0
sympy==1.14.0
tokenizers==0.22.1
torch==2.9.0
tqdm==4.67.1
transformers==4.57.1
triton==3.5.0
typing_extensions==4.15.0
urllib3==2.5.0
5

I’m working on a hackathon with MedGemma and I’m running in this issue as well: GitHub - ODSCGoogleHackhathon/googol at feat/medgemma

I don’t think we can make the model much light weight with the requirements in general. Here is the requirements I’m using for the model:

```txt

pydantic==2.10.5

pydantic-settings==2.7.0

python-dotenv==1.0.1

# Image Processing

Pillow==11.0.0

# Utilities

aiofiles==24.1.0

httpx==0.28.1

# MedGemma - Heavy ML Stack

transformers>=4.45.0

torch>=2.0.0
```