Currently the plugin crashes when running certain models (see #27, #28, #35, and #39) -- this PR fixes those model parsing errors.
There are 3 main parts to the fixes (roughly divided by commit):
5a8b2454549ed17928a4e714395a929c8702cc89: Use encode() when encode_amortized() is unavailable in the model
c17a814783df769932999588b1cb5b84b9548b7d: Check whether model supports stereo -- if not, send mono instead of stereo input into decode(). This enables support for mono models.
35450abc23a8df7e9c05ea82c4e9a7171b79da7e: Check top_level named_attributes if available (instead of named_buffers within _rave). This enables support for RAVE v2 models.
Tested successfully with all the models currently on the RAVE models download page. However, the darbouka_onnx model doesn't sound quite correct when run -- not sure whether this is due to the model being v2 or ONNX (will continue investigating), but at the very least the plugin should no longer crash.
Currently the plugin crashes when running certain models (see #27, #28, #35, and #39) -- this PR fixes those model parsing errors.
There are 3 main parts to the fixes (roughly divided by commit):
encode()
whenencode_amortized()
is unavailable in the modeldecode()
. This enables support for mono models.named_attributes
if available (instead ofnamed_buffers
within_rave
). This enables support for RAVE v2 models.Tested successfully with all the models currently on the RAVE models download page. However, the
darbouka_onnx
model doesn't sound quite correct when run -- not sure whether this is due to the model being v2 or ONNX (will continue investigating), but at the very least the plugin should no longer crash.