jaxfne
jaxfne is a compact JAX package for Tensor-Field Neural Equations (TFNE): a typed computational chain from neural emitters to source tensors, field-proxy operators, probe readouts, objective reports, optimizers, and run manifests.
Install
pip install jaxfne
Visualization extras:
pip install "jaxfne[viz]"
Development checkout:
git clone https://github.com/HNXJ/jaxfne.git
cd jaxfne
pip install -e ".[dev,viz]"
Canonical import:
import jaxfne as jtfne
Object grammar
Every jaxfne program is the same linear chain. Each step returns the input to the next, so the whole pipeline reads as one fluent sequence:
setup -> config -> construct -> simulate -> visualize -> tune/objective -> optimize -> export
enable_x64 Configuration Model Signals vis.* Objective Model.tune manifest / save_*
The full typed chain underneath that sequence:
Config
-> Runtime(seed, dtype, backend, jit, vmap, duration_ms, dt_ms)
-> Identity(area, layer, cell_type, unit_id, position)
-> Emitter(theta_e, state_0, drive, noise, key)
-> SourceMap(source_mode, source_calibration_status, support, normalization)
-> FieldProxy(kernel, geometry_metadata, field_solver_status, field_claim_level)
-> Probe(kind, selector, channel_geometry, units_status, method)
-> Signals(spk, vm, source, lfp_proxy, csd_proxy, eeg_proxy, meg_proxy, spectrolaminar_proxy, emm_proxy)
-> Objective(metrics, targets, gates, nulls, rejection_reasons)
-> Optimizer(search_space, budget, key, constraints)
-> Manifest(run_id, version, repo_sha, runtime_report, artifact_paths, asset_hashes, truth_gates)
-> Validation(finite_outputs, strict_json, png_assets, notebook_receipts, optional_dependency_laziness)
Minimal workflow
import jaxfne as jtfne
jtfne.enable_x64() # setup: x64 before arrays
cfg = jtfne.build_laminar_column() # config: V1, n=1000, flat E:I (legacy default)
cfg = (cfg.set_emitter("izhikevich", "cortical_eig")
.probes(["spikes", "V_m", "LFP", "CSD"], n_contacts=16)
.field(domain="laminar_column", conductivity="proxy", boundary="mean_zero_neumann"))
model = jtfne.construct(cfg) # construct: Configuration -> Model
signals = jtfne.simulate(model, duration_ms=1000.0, dt_ms=0.5, seed=0) # simulate -> Signals
spk = signals.get("spk") # (n_steps, n_neurons) spike raster
vm_e = signals.get("vm", cell_type="E") # membrane voltage for E cells
Canonical cortex (default prior)
The verified ground-truth laminar prior is a first-class API surface — no source editing required. Excitatory fraction rises with depth (L6 ≈90% E), inhibition peaks superficially (L1 50% I, no PV), PV concentrates at L4; overall ≈77E:23I.
# Canonical laminar cortex: real per-layer composition + laminar placement.
cfg = jtfne.build_laminar_column(n=1000, ei_profile="canonical")
# Multi-area hierarchy with the canonical prior in each area:
cfg = jtfne.build_multi_area_columns(["V1", "V4", "PFC"], ei_profile="canonical")
ei_profile="flat" (the default) preserves the legacy depth-invariant
composition and uniform3d placement unchanged; ei_profile="canonical"
auto-routes to laminar placement so each neuron keeps its layer label and the
per-layer E:I gradient is expressed. The exported constants
jtfne.CANONICAL_LAYER_CELL_TYPE_FRACTIONS, jtfne.CANONICAL_Z_BANDS, and
jtfne.DEFAULT_LAYERS document the prior directly.
Tune toward a target
obj = jtfne.rate_synchrony_targets() # defaults: 10 Hz, kappa 0 (async-irregular)
result = model.tune(obj, optimizer="AGSDR", steps=50)
tuned = result.model
manifest = jtfne.manifest(cfg, signals=signals) # export: strict JSON-safe run manifest
Checkout validation
python -m compileall -q jaxfne tests examples scripts
PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 PYTHONPATH=. python -m pytest tests -q --tb=short
PYTHONPATH=. python scripts/audit_notebooks_and_assets.py --check
PYTHONPATH=. python scripts/audit_notebook_grammar.py --check
mkdocs build --strict
1 maintainer
Release files for jaxfne 0.4.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| jaxfne-0.4.1.tar.gz | 8.2 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| jaxfne-0.4.1-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 8.5 MB
Release files / jaxfne-0.4.1.tar.gz
| Download URL | jaxfne-0.4.1.tar.gz |
|---|---|
| Size | 8.2 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
1600d962c8107a742d72202184270ba9c3a26ec26c299dd9d65d820bdf2cd6d4
|
|
BLAKE2b-256 checksum How to use checksums |
230ed726c828303675305abc05f40b2188b3f745af112c72bcb32da2db5c2822
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 21, 2026.
Transparency logRelease files / jaxfne-0.4.1-py3-none-any.whl
| Download URL | jaxfne-0.4.1-py3-none-any.whl |
|---|---|
| Size | 276.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
6266ec3a28f1b833ab4839e4d53a104afd33b9c0e8023e29123876040d9e89ed
|
|
BLAKE2b-256 checksum How to use checksums |
5c3db16ac52d18b14d6ce94b2c8bd95142c2b44409bca1e517bd7bed72cd5bac
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Jun 21, 2026.
Transparency log
