Skip to content

Conversation

@sdatkinson
Copy link
Owner

Resolves #148

- Refactored DSP class to eliminate channel-specific input/output level methods, consolidating them into single methods for input and output levels.
- Updated related methods and tests to reflect the new simplified interface, ensuring backward compatibility.
- Adjusted buffer management in the Buffer class to accommodate the changes in DSP level handling, ensuring all channels use consistent buffer sizes.
- Updated the script to remove all contents from the build directory instead of only untracked files, streamlining the cleanup process before running benchmarks.
- Included multiple LSTM test cases to validate various functionalities, including basic operations, handling of different buffer sizes, and state evolution.
- Ensured comprehensive coverage for LSTM layers by adding tests for multichannel inputs and large hidden sizes.
…onstants

- Updated multiple ConvNet test functions to replace hardcoded input and output channel values with defined constants for better readability and maintainability.
- Ensured consistency across tests by using the same channel definitions in all relevant instances.
…annels

- Implemented a new test case to verify that the process method does not allocate or free memory when handling 3 input channels and 2 output channels.
- The test includes setup for weights and input/output buffers, ensuring the output is valid across various buffer sizes.
- This enhances the coverage of WaveNet functionality in real-time processing scenarios.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[BREAKING] Multi-channel (multi-input, multi-output) models

2 participants