add ascend 910B npu support #24
Conversation
|
感谢感谢，等我晚点并入，发版 |
There was a problem hiding this comment.
Pull Request Overview
Adds support for Ascend 910B NPU by integrating the ONNXRuntime CANN execution provider.
- Introduces a new
CANNExecutionProviderenum and related configuration flag - Extends session setup to conditionally prepend the CANN EP with its specific options
- Updates the main entrypoint to accept and forward a
use_cannargument
Reviewed Changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| rapid_layout/utils/infer_engine.py | Added CANN_EP enum, cfg_use_cann handling, _check_cann and EP insertion logic |
| rapid_layout/main.py | Exposed new use_cann parameter and passed it into OrtInferSession configuration |
Comments suppressed due to low confidence (2)
rapid_layout/main.py:54
- No tests cover the new
use_canncode path; consider adding unit or integration tests to verify the CANNExecutionProvider logic.
use_cann: bool = False,
| @@ -51,6 +51,7 @@ def __init__( | |||
| iou_thres: float = 0.5, | |||
| use_cuda: bool = False, | |||
| use_dml: bool = False, | |||
| use_cann: bool = False, | |||
There was a problem hiding this comment.
[nitpick] The new use_cann parameter is not documented in the function docstring or CLI help; consider updating documentation to describe this option.
Copilot uses AI. Check for mistakes.
| @@ -71,6 +73,16 @@ def _get_ep_list(self) -> List[Tuple[str, Dict[str, Any]]]: | |||
| } | |||
| EP_list = [(EP.CPU_EP.value, cpu_provider_opts)] | |||
|
|
|||
| self.use_cann = self._check_cann() | |||
There was a problem hiding this comment.
[nitpick] Provider initialization follows a similar pattern across _check_* methods and EP_list insertion; consider refactoring common logic into a helper to reduce duplication.
Copilot uses AI. Check for mistakes.
910B的支持，已在CANN8通过测试，需要依赖onnxruntime-cann