Thank you very much for this amazing tool! It really helps a lot! In the newest version of the Docker setup, I have this error:

When attempting to use point or scribble interactions on the SlicerNNInteractive server, operations fail with a CUDA error when creating tensors using the Docker setup.

Environment Information

• Docker Image: coendevente/nninteractive-slicer-server:latest
• CUDA Version: 12.1.0
• Python Version: 3.12 (conda environment: nnInteractive)
• GPU: Full GPU access (not virtualized/partial)
• Input Image Dimensions: 797×512×512 (from server logs)
• Operating System: Linux (Jetstream cloud instance)

Detailed Error

The server fails with RuntimeError: CUDA error: invalid argument when trying to create a CUDA tensor with torch.zeros() in the _initialize_interactions method. This happens after successfully uploading an image and attempting to add interactive prompts.

Steps to Reproduce

1. Start the Docker container: docker run --gpus all --rm -it -p 1527:1527 coendevente/nninteractive-slicer-server:latest
2. Connect to the server from 3D Slicer using the nnInteractive extension
3. Upload an image (in my case, with dimensions 797×512×512)
4. Try to add a point or scribble interaction
5. Server crashes with the CUDA error shown in the stack trace

Error Stack Trace

RuntimeError: CUDA error: invalid argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

The error specifically occurs in:

File "/opt/conda/envs/nnInteractive/lib/python3.12/site-packages/nnInteractive/inference/inference_session.py", line 140, in _initialize_interactions
  self.interactions = torch.zeros(