fix: auto-upgrade model opset to 21 for int16/uint16 QDQ quantization#28202

Merged

tianleiwu merged 7 commits into

microsoft:mainfrom

Rishi-Dave:rishidave/feat/quant-utils-opset-bump-16bit

May 14, 2026

Commits on Apr 23, 2026

fix: auto-upgrade model opset to 21 for int16/uint16 QDQ quantization
Rishi-Dave
committed

Commits on May 1, 2026

fix(quantization): bump opset for int16 activations too
Rishi-Dave
committed

Commits on May 3, 2026

fix(quantization): do not pre-set UseQDQContribOps for int16 types in get_qdq_config
Rishi-Dave
committed

Commits on May 4, 2026

fix(quantization): scan TensorQuantOverrides for 16-bit and recompute UseQDQContribOps after opset bump
Rishi-Dave
committed

Commits on May 5, 2026

fix: harden override scan and update tests for opset-21 bump
Rishi-Dave
committed

Commits on May 11, 2026

Log debug message instead of silent pass on malformed overrides
Rishi-Dave
committed

Commits on May 12, 2026

fix(quant): guard extra_options None and strengthen 16-bit opset tests
Rishi-Dave
committed