CRISPI

CRISPI

If you are using a different version of CUDA/Jetpack do the followings to run the real-time trace extraction: Here I'm trying to run it on Jetpack 5.1.1. Use a CuPy build for CUDA 11.x instead of 12.x. Do this: Use the stimviewer env’s Python to avoid activating shells PY=/media/aharonilabjetson2/NVMe/mamba_root/envs/stimviewer/bin/python

$PY -m pip uninstall -y cupy-cuda12x cupy Try wheel for CUDA 11.x (works on many Jetsons) $PY -m pip install cupy-cuda11x==13.5.1

If wheel not available on your aarch64, build from source (long compile): $PY -m pip install cupy==13.5.1 --no-binary cupy

Verify using: $PY - << 'PY' import cupy as cp, cupy.cuda.runtime as r print('devices:', r.getDeviceCount()) cp.arange(1); print('CuPy OK') PY Also update your pins: In STIMViewer_CRISPI/requirements.txt: use cupy-cuda11x==13.5.1 In STIMViewer_CRISPI/environment.yml: replace cupy-cuda12x with cupy-cuda11x Once CuPy matches CUDA 11.4, CUDA stays enabled; the runtime guard will pass and GPU paths run.

---

Jetson.GPIO isn’t installed in your current Python environment (stimviewer). Install it there and give your user GPIO access, then retry the Test Pulse. Do this: In stimviewer venv: pip install Jetson.GPIO

Ensure your user can access GPIO without sudo: sudo usermod -aG gpio $USER sudo udevadm control --reload-rules && sudo udevadm trigger log out/in (or reboot) to apply group change

Optional (if pip fails), install the system package: sudo apt-get update sudo apt-get install -y python3-jetson-gpio

---

Kill any stray senders:

pgrep -fa zmq_mask_sender.py pkill -f zmq_mask_sender.py