| refactor: implement async edge proxy for distributed inference
- Decouple monolithic inference into pure Python Pre, Infer, and
Post core components.
- Introduce InferenceCoordinator for zero-copy single-process inference.
- Refactor lerobot_policy_node into an Asynchronous Edge Proxy to
support Edge-Cloud split without breaking action_dispatch.
- Remove redundant base nodes (base_model_node, passive_inference_node)
and legacy component wrappers.
- Add pure_inference_node and cloud_inference.launch.py.
- Support runtime mode switching via robot_config YAML.
Signed-off-by: XiaoqiangWu <wuxiaoqiang.rtos@huawei.com>
| 2 个月前 |
| inference_service: honor runtime policy device
Route LeRobot config loading through a runtime policy copy so launch device settings win.
Keep source models unchanged and clean temporary policy copies after upstream loaders use them.
Add startup diagnostics for model config device, runtime backend, and tensor device.
Cover missing device fields and compiled backend config handling in regressions.
Signed-off-by: Shi Xin <shixin21@h-partners.com>
| 6 天前 |
| inference_service: generalize compiled wrappers
Split compiled inference into adapter and runtime layers.
Keep backend execution separate from model-family semantics.
Resolve Ascend OM artifacts through config.om.json.
Generate config.om.json from the ATC export path.
Expose backend_type beside policy_type for diagnostics.
Signed-off-by: XiaoqiangWu <wuxiaoqiang.rtos@huawei.com>
| 8 天前 |
| inference_service: generalize compiled wrappers
Split compiled inference into adapter and runtime layers.
Keep backend execution separate from model-family semantics.
Resolve Ascend OM artifacts through config.om.json.
Generate config.om.json from the ATC export path.
Expose backend_type beside policy_type for diagnostics.
Signed-off-by: XiaoqiangWu <wuxiaoqiang.rtos@huawei.com>
| 8 天前 |
| inference_service: honor runtime policy device
Route LeRobot config loading through a runtime policy copy so launch device settings win.
Keep source models unchanged and clean temporary policy copies after upstream loaders use them.
Add startup diagnostics for model config device, runtime backend, and tensor device.
Cover missing device fields and compiled backend config handling in regressions.
Signed-off-by: Shi Xin <shixin21@h-partners.com>
| 6 天前 |
| inference_service: generalize compiled wrappers
Split compiled inference into adapter and runtime layers.
Keep backend execution separate from model-family semantics.
Resolve Ascend OM artifacts through config.om.json.
Generate config.om.json from the ATC export path.
Expose backend_type beside policy_type for diagnostics.
Signed-off-by: XiaoqiangWu <wuxiaoqiang.rtos@huawei.com>
| 8 天前 |
| inference_service: generalize compiled wrappers
Split compiled inference into adapter and runtime layers.
Keep backend execution separate from model-family semantics.
Resolve Ascend OM artifacts through config.om.json.
Generate config.om.json from the ATC export path.
Expose backend_type beside policy_type for diagnostics.
Signed-off-by: XiaoqiangWu <wuxiaoqiang.rtos@huawei.com>
| 8 天前 |