Sinabs

Latest version: v2.0.0

Safety actively analyzes 681874 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 3 of 6

1.1.5

------

* Fix backward pass for MaxSpike
* update ChangeLog after add support of speck2e
* add 'spiking' and 'parameter' subdictionaries to layer stats in snn analyzer
* add reset method for analyzer
* SNNAnalyzer able of tracking firing rates across batches
* make sure synops are accumulated across batches
* add speck2e devkit and testboard support
* remove redundant comment code
* Device input graph stopped on destruction of the DynapcnnNetwork object
* Writing all v mem to zeros changed. -> Now when you call this method it creates a temporary buffer and a temporary graph, starts the graph, writes the events, stops the graph then it is destroyed
* Implement an input buffer for all devices: -> In the config\_builder there is now a function called get\_input\_buffer which gives a source node. -> For each chip this is implemented to get the appropriate source node. -> To method changed to construct the input buffer. -> Forward method changed to write the events to this source node
* improve the plotting of which metrics are tracked in the tutorial notebook
* add a plot for firing rate histograms in the snn synops tutorial
* try using reshape instead of unflatten to be compatible with pytorch 1.8
* attempt to make it pytorch 1.8 compatible and fix synops training tutorial
* replace SNNSynopCounter with SNNAnalyser, which calculates statistics for both param and spiking layers
* Name added to io
* naive copy of config generation
* moved kill bits to a separate method
* Type definition updated to Tuple
* compatibility with torch 1.13
* Name added to io
* Failing unit test
* Requirements for samna version updated
* Change to get the same functionality in 'to()' method in samna Version: 0.21.2.0

1.1.4

------

* autoformat all docstrings using docformatter
* fix bug where added\_spike\_output is a ReLU
* Update README.md
* Update ci-pipeline.yml
* Update ci-pipeline.yml
* Update ci-pipeline.yml

1.1.3

------

* Update ci-pipeline.yml
* add generic reset\_states and zero\_grad utility functions

1.1.2

------

* make sure that add\_spike\_output doesn't add spiking layer in original ANN
* removed unused imports
* sort all imports automatically
* moved kill bits to a separate method
* Enable top-level conversion of sinabs layers with 'replace\_module' function. Resolves issue 60
* Enable converting sinabs layers with
* add documentation about how to release a new Sinabs version
* Hotfix new arg\_dict

1.1.1

------

* add arg\_dict property to StatefulLayer
* Update ci-pipeline.yml
* Update ci-pipeline.yml
* Update ci-pipeline.yml
* bump Python version under test to 3.10
* get rid of exodus import in test\_conversion
* deprecated 'backend' parameter in from\_torch
* add a replace\_module function that will replace specific layer according to a mapper\_fn. Reworked from\_torch
* requirement bumped to 1.0.7

1.0.7

------

* check for pytorch version under test and either call testing.assert\_close or assert\_allclose depending on the version
* using torch.equal instead of np equal
* tests fixed
* fixing test

Page 3 of 6

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.