Installation¶
git clone https://github.com/SPICExLAB/IMU-Bench && cd IMU-Bench
python -m venv imu-bench && source imu-bench/bin/activate
pip install -e .
imu-bench tasks
Python 3.9 or newer and one GPU. Install the CUDA build of PyTorch first if pip picks the CPU wheel for your
platform; every run in the paper used one NVIDIA A100 or H100.
The install is editable because the harness runs inside the clone:
| Path | Holds |
|---|---|
configs/ |
task templates and pretraining configs |
bench/registry/ |
task cards, encoder cards, protocols, tracks |
bench/generated/configs/ |
the generated config of every cell (the paper's are committed) |
runs/ |
training runs: checkpoints and logs, one directory per cell |
bench/results/ |
one result JSON per cell |
SLURM
imu-bench eval --slurm submits one train → score chain per cell. The generated scripts activate imu-bench/
by default; set IMUBENCH_VENV to use another environment directory, and BENCH_PARTITION / BENCH_ACCOUNT
for your queue.
Next: Quickstart, then Datasets.