Some fatal bugs and when the bug is fixed are shown in this table. Note that the bug fixed after a release indicates that this bug may exist in the release.
| Bugs/Releases | Date |
|---|---|
| Release: 0.0.0.0.4 | 2021-03-25 |
| Release: 0.0.0.0.6 | 2021-07-03 |
| Bug: Tempotron, #88. This bug makes the maximum value of psp kernel not be v_threshold. But considering that weights are learnable, this bug may not have much influence. | 2021-07-17 |
| Bug: SpikingRNNBase, #101. This bug makes SpikingVanillaRNN, and SpikingGRU use SpikingLSTMCell in their layers. | 2021-08-26 |
| Bug: Cupy backend for spiking neurons, #106. This bug makes spiking neurons with cupy backend output wrong spikes and voltages. This bug has no influence on release 0.0.0.0.4, which does not use cupy. | 2021-09-16 |
| Release: 0.0.0.0.8 | 2021-11-21 |
| Bug: MultiStepParametricLIFNode, #151. This bug makes the gradient of the learnable parameter in MultiStepParametricLIFNode incomplete when backend is cupy. | 2021-12-10 |
| Release: 0.0.0.0.10 | 2022-03-04 |
Bug: When using CuPy with version >= 10, CuPy will change torch.cuda.current_device() to 0, cupy/cupy#6569. This bug will break training when using Distributed Data Parallel (DDP). |
2022-03-22 |
| Release: 0.0.0.0.12 | 2022-05-01 |
| Release: 0.0.0.0.14 | 2023-03-20 |
Bug: The SHD dataset can not be integrated to frames by fixed duration correctly, #361. This bug will create wrong frames if the user uses spikingjelly.datasets.SpikingHeidelbergDigits(duration=duration, ...) when duration is not None. |
2023-04-01 |
Bug: TemporalWiseAttention and MultiDimensionalAttention are not callable, #363. The user will encounter the ... object is not callable error. |
2023-04-01 |