Skip to content

MLECO-6696: Using trained Conformer model

Kshitij Sisodia requested to merge feature/mleco-6696-trained-conformer into experimental/executorch

Adding provision to use a trained Conformer (PyTorch) model for ASR use case. ExecuTorch resource class now contains an option for providing a "lowering" script which can be used instead of the default AOT Arm compiler script provided by Arm backend in ExecuTorch sources.

Also, with this change, Python virtual environment setup will install executorch==1.0.0 by default. Only if this is not present will we fall back to installing it from source.

NOTE: default PyPi package may not work for Linux aarch64.

Change-Id: I0cf5eee79214ffb0a58642c8e90d301e8b1dcf7b Signed-off-by: Kshitij Sisodia kshitij.sisodia@arm.com

Merge request reports