Improve Metal backend with hybrid CPU+GPU+ANE inference for macOS#1148

Open

ChinChangYang wants to merge 27 commits intolightvector:masterfrom

ChinChangYang:coreml-backend

Commits on Jan 1, 2026

Add configurable FP16/FP32 precision to CoreML backend

ChinChangYang
and
claude
committed

Commits on Jan 6, 2026

Add CoreML backend CI job to GitHub Actions workflow

ChinChangYang
and
claude
committed

Commits on Jan 20, 2026

Simplify CoreML model loading with dynamic batch size support

ChinChangYang
and
claude
committed
Add FP32 GPU-only mode using MPSGraph to bypass CoreML converter

ChinChangYang
and
claude
committed

Commits on Jan 21, 2026

Commits on Jan 24, 2026

Unify CoreML and Metal into single Metal backend

ChinChangYang
and
claude
committed

Commits on Feb 26, 2026

Refactor MPSGraph inference to use explicit command buffer

ChinChangYang
and
claude
committed

Commits on Feb 27, 2026