Skip to content

feat(detection): route Apple Silicon YOLO to Neural Engine (NPU)#142

Merged
solderzzc merged 1 commit intodevelopfrom
feature/apple-npu-detection
Mar 10, 2026
Merged

feat(detection): route Apple Silicon YOLO to Neural Engine (NPU)#142
solderzzc merged 1 commit intodevelopfrom
feature/apple-npu-detection

Conversation

@solderzzc
Copy link
Member

On Apple Silicon, force CoreML compute_units to CPU_AND_NE so object detection runs on the 16-core Neural Engine instead of GPU, leaving GPU free for LLM/VLM inference.

Changes:

  • env_config.py: add compute_units field to BackendSpec, MPS defaults to cpu_and_ne, monkey-patch coremltools.MLModel during YOLO load
  • detect.py: report compute_units in ready event
  • SKILL.md: add compute_units parameter, update acceleration table
  • test_env_config_ane.py: 15 unit tests for NPU logic

On Apple Silicon, force CoreML compute_units to CPU_AND_NE so object
detection runs on the 16-core Neural Engine instead of GPU, leaving
GPU free for LLM/VLM inference.

Changes:
- env_config.py: add compute_units field to BackendSpec, MPS defaults
  to cpu_and_ne, monkey-patch coremltools.MLModel during YOLO load
- detect.py: report compute_units in ready event
- SKILL.md: add compute_units parameter, update acceleration table
- test_env_config_ane.py: 15 unit tests for NPU logic
@solderzzc solderzzc changed the base branch from master to develop March 10, 2026 06:21
@solderzzc solderzzc merged commit f730030 into develop Mar 10, 2026
1 of 2 checks passed
@solderzzc solderzzc deleted the feature/apple-npu-detection branch March 10, 2026 06:48
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant