4. Optimize training performance: torch.compile, AMP, persistent workers - #9

Merged

Adithya-Thonse merged 5 commits into

TexasInstruments:mainfrom

musicalplatypus:pr/performance-optimizations

Jul 28, 2026

Commits on Jul 28, 2026

Optimize training performance: torch.compile, AMP, persistent workers, eval efficiency

t5fkg8d44d-beep
and
claude
committed
Thread performance flags (compile_model, native_amp) through modelmaker config pipeline

t5fkg8d44d-beep
and
claude
committed
Fix two peer-review regressions in performance PR

t5fkg8d44d-beep
and
claude
committed
fix: remove dead confusion_matrix_total init in evaluate_classification

t5fkg8d44d-beep
and
claude
committed
fix: use squeeze(-1) to avoid collapsing batch dim in evaluate_classification

t5fkg8d44d-beep
and
claude
committed