| inference_service: add Ascend OM backends
Move the existing ACT OMModel and ACT wrappers from the LeRobot patch
stack into inference_service so issue #48 can be resolved without
carrying those patches in the submodule.
Keep CPU, CUDA, and NPU loading unchanged. Select the OM adapters
only when the explicit ascend_om or ascend_om_3403 backend is requested.
Delete the migrated LeRobot patches from the active patch lists and
keep tests covering backend selection and patch filtering.
Signed-off-by: Shi Xin <shixin21@h-partners.com>
| 24 天前 |
