Frequently Asked Questions¶
Installation and setup¶
Q: What Python version does jaxfne require?
A: jaxfne requires Python 3.10 or later. Most examples and tests use Python 3.11 or 3.13.
Q: Can I run jaxfne on GPU?
A: Yes. jaxfne uses JAX natively, which supports GPU execution. Set JAX device configuration as needed. CPU-first examples validate correctness; GPU execution is optional.
Q: How do I install JAX with GPU support?
A: See the JAX installation guide. Install jax with [cuda] or [tpu] extras if using accelerators.
Basic usage¶
Q: What's the simplest way to use jaxfne?
A: See Quickstart for a minimal example. It takes ~5 lines of code to configure, build, and simulate a 100-neuron network.
Q: Can I use jaxfne with Jaxley models?
A: Yes. See Jaxley interoperability for how to mount Jaxley-style outputs as source tensors.
Q: What readout operators are available?
A: Eight operators: SPK (spikes), Vm (voltage), source, LFP-proxy, CSD-proxy, EEG-proxy, MEG-proxy, EMM-proxy. See Probe operators.
Advanced workflows¶
Q: How do I calibrate outputs to physical units?
A: Default readouts are computational proxies. Calibration requires empirical data, geometry specifications, and solver validation. See Calibration for the calibration-ready design, and Limitations and future plans for the declared field regimes.
Q: What metadata does jaxfne attach to outputs?
A: All outputs include JSON-safe manifests with operator status, units, assumptions, and validation metadata. See Output bundles.
Q: Can I use custom emitter models?
A: Yes. Implement the emitter interface and pass outputs as JAX arrays. See Emitters for details.
Troubleshooting¶
Q: I get a JAX device error. What's wrong?
A: Check JAX device availability with jax.devices(). jaxfne is CPU-safe by default. If using GPU, verify CUDA/GPU setup via JAX documentation.
Q: How do I reproduce results?
A: Use explicit PRNG seeds in simulations. jaxfne uses deterministic JAX operations; same seed → same trajectory.
Q: Where can I ask more questions?
A: Open an issue or discussion on GitHub.