Skip to content

Version 2.8.0

Latest
Latest

Choose a tag to compare

View all tags
@chaoming0625 chaoming0625 released this 19 Jun 11:34
V2.8.0
b6b5cc5
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Release Date: June 19, 2026

Version 2.8.0 is a broad correctness, quality, and maintainability release. It bundles an audited, library-wide bug-fix sweep — neuron and synapse dynamics, ODE/SDE/FDE integrators, the math and object-transform layer, dnn layers, optimizers and losses, analysis tooling, and the simulation runners — each accompanied by new regression tests. Alongside the fixes, the release introduces static typing with a mypy CI gate, raises test coverage from roughly 84% to over 92%, co-locates tests with the modules they cover, and removes duplicated internals by reusing the shared braintools and brainstate implementations. Documentation and the bundled examples were repaired and refreshed.

Highlights

  • Audited correctness sweep across the whole library, each fix backed by regression tests.
  • Static typing: type annotations for core utilities, PEP 561 (py.typed), and a new mypy continuous-integration gate (#834).
  • Test-suite overhaul: coverage raised from ~84% to 92%+ (#836), with tests co-located as <module>_test.py (#832).
  • Reduced duplication: reuse braintools (init, metric, surrogate) and brainstate transforms instead of maintaining forked copies (#835, #833, #831).

Bug Fixes

Neuron and synapse dynamics

  • CondNeuGroup: corrected synaptic-current scaling that attenuated currents by ~1000× (#842).
  • DualExp: correct handling of equal time constants, plus per-neuron STP reset (#847).
  • dynold: fixed STP construction, sparse-synapse drift, plasticity decay, and Bellec initialization (#848).
  • Rate models: fixed RNNCell.reset_state crash and ThresholdLinearModel noise dt-scaling (#850).

Integrators

  • RKF45 node handling and Kl/Pl SRK diffusion weights (#841).
  • Fractional integrators: CaputoEuler.reset state desynchronization and set_default_fdeint (#846).
  • Prevented a MilsteinGradFree out-of-memory crash (#837).

Math, sparse, and object transforms

  • coo_to_csr bounds checking, plus a sparse/event/delay regression suite (#845).
  • Compatibility fixes: gelu integer input, unflatten negative dimension, segment_mean on Array (#844).
  • Object transforms: cond/ifelse, collectors, and VariableView edge cases (#840).
  • ShardedArray pytree flattening and remove_diag guard (#839).

Neural-network layers

  • BatchNorm running-variance bias, pooling channel_axis, and LayerNorm error handling (#843).

Optimizers and losses

  • Audited correctness fixes for Adan, SM3, and several loss functions (#838).

Analysis

  • Fixed-point classification, nullcline selection, gradient-descent batching, and plotting keyword handling (#849).

Runtime and runners

  • DSRunner: memory_efficient output/monitors and eager bound-validation checks (#851).
  • IntegratorRunner: dict-monitor regression (#854).
  • Runners: jit-dict mutation, dict string monitors, and multi-device parallel concatenation (#852).
  • Inputs / algorithms / connectivity: deprecation aliases, regression fits, CSR guard, and dtype handling (#853).
  • Earlier audited correctness, API-drift, and edge-case fixes with a regression and coverage suite (#830).

Quality and Tooling

  • Added typing for core utilities, enabled PEP 561, and added a mypy CI job (#834).
  • Comprehensive coverage tests, raising coverage from ~84% to 92%+ (#836).
  • Co-located tests as <module>_test.py and removed the legacy tests/ trees (#832).
  • Reused braintools.init / braintools.metric in brainpy.initialize and losses (#835).
  • Reused braintools.surrogate and removed the duplicate brainpy.math.surrogate (#833).
  • CI: replaced remove_vmap with brainstate.transform.unvmap and resolved brainstate 0.5.1 API drift (#831); green-lighted the Dense fit-flag tracer, buffer-donation test pollution, and L1 loss contract, and restored the Codecov token (#855).

Documentation and Examples

  • Converted public-API docstrings to NumPy style and enforced the convention (#856).
  • Repaired documentation notebooks and fixed Hz-input and RNNCell.reset(batch_size) bugs (#857).
  • Repaired six broken example scripts — API drift, data paths, and channel handling (#858).
  • Updated documentation links and the Python-version requirement in README.md; added .gitattributes for line endings.
  • Served the project logo as WebP from brainx.chaobrain.com.

Dependencies

  • Bumped sphinx to >=9.0.4,<9.1.0 (#827), sphinx-book-theme to >=1.2.0 (#826), and jupyter-sphinx to >=0.5.3 (#825).
Assets 2
Loading