sdatkinson/neural-amp-modeler v0.13.0

Version 0.13.0

on GitHub

What's Changed

• [FEATURE] Sequential compositional models by @sdatkinson in #596
• [FEATURE] Support loading WAVE files in 32-bit floating-point by @sdatkinson in #599
• Add build status badge to README by @Khalian in #601
• [FEATURE] New activations by @sdatkinson in #625
• [FEATURE] WaveNet bottlenecks by @sdatkinson in #626
• [FEATURE] WaveNet head 1x1 layer by @sdatkinson in #627
• [FEATURE] WaveNet: Add layer1x1 config with grouped convolutions, able to be inactive by @sdatkinson in #628
• [FEATURE] WaveNet: FiLMs, grouped convolutions by @sdatkinson in #629
• [FEATURE] WaveNet: Support conditioning submodule (condition_dsp) by @sdatkinson in #630
• [BUGFIX] Fix bugs with WaveNet receptive field and tensor shapes when a condition_dsp module is present by @sdatkinson in #632
• [BUGFIX] Fix graceful shutdown not generating model (Issue #501) by @sdatkinson in #612
• [BUGFIX] Remove "upgrade" button from GUI trainer, replace with a modal alerting user of new version by @sdatkinson in #633
• Integration tests by @sdatkinson in #634
• [FEATURE] Support for slimmable WaveNets via channel-slicing. by @sdatkinson in #636
• [BUGFIX] Fix WaveNet Layer.bottleneck when using pairing activations by @sdatkinson in #640
• [FEATURE] Support slimming via channel slicing only at specific allowed channel counts by @sdatkinson in #639
• [FEATURE] "Boosting mode" for slimmable WaveNet (channel-slicing) by @sdatkinson in #641
• [FEATURE] Initialization strategies "smallest_and_zeros" and "channel_causal" for channel-slicing slimmable WaveNet by @sdatkinson in #642
• [BUGFIX] slimmable WaveNet: Export allowed channels by @sdatkinson in #643
• Rework implementation of boosting for channel-slicing slimmable WaveNet, update model export version to 0.7.0. by @sdatkinson in #644
• Validate slimmable configuration by @sdatkinson in #646
• [FEATURE,BREAKING] WaveNet: Support different kernel sizes on different layers of a layer array by @sdatkinson in #648
• [BREAKING,FEATURE] WaveNet: Support for model head, layer array head_rechannel with kernel size > 1 by @sdatkinson in #653
• [SECURITY] Update pytorch_lightning dependency to exclude specific versions by @sdatkinson in #657
• [FEATURE] Packed WaveNet training by @sdatkinson in #659
• Clarify packed training slimmable docs by @sdatkinson in #660
• Update simplified trainers to use A2 by @sdatkinson in #661
• [FEATURE] Support hooks post model dict formation in .export() by @sdatkinson in #663
• [BUGFIX] Add model-data handshake to core training routine by @sdatkinson in #664
• [FEATURE] Hooks that act jointly on train and validation datasets by @sdatkinson in #665
• Compensate metadata.loudness when _ScaleOutputHook scales head_scale by @woodybury in #668
• Address PR 668 review nits by @sdatkinson in #670
• Log per-submodel MRSTFT metrics during packed validation by @woodybury in #669

New Contributors

• @Khalian made their first contribution in #601

Full Changelog: v0.12.2...v0.13.0

Support

A2 was developed in partnership with TONE3000, who contributed research, engineering, and funding. A2 models can be integrated into apps, plugins, and hardware via the TONE3000 API.