Add ONNX Runtime backend with multiple execution providers#1164

Draft

ChinChangYang wants to merge 22 commits intolightvector:masterfrom

ChinChangYang:onnx-backend

Commits on Feb 28, 2026

Add ONNX Runtime backend with configurable execution providers

ChinChangYang
and
claude
committed

Commits on Mar 1, 2026

Support humanv0 model with SGF metadata encoder in ONNX backend

ChinChangYang
and
claude
committed
Simplify ONNX backend: use MatMul, add null guard, hoist variable

ChinChangYang
and
claude
committed
Support loading raw .onnx model files in ONNX backend

ChinChangYang
and
claude
committed
Fix tellg() failure check and avoid large model copy in ONNX backend

ChinChangYang
and
claude
committed
Fix thread_local misuse and dead code in ONNX backend

ChinChangYang
and
claude
committed
Improve ONNX backend robustness and defaults

ChinChangYang
and
claude
committed

Commits on Mar 3, 2026