feat(perf): support composite model generation benchmarking#1059

Open

xieofxie wants to merge 13 commits into

mainfrom

hualxie/infer_automodel

Commits on Jul 6, 2026

Improve GenAI session: smarter max_length, DML skip, flat-decoder support, and response logging (#1054)
Hualiang Xie
committed
feat(perf): support composite model generation benchmarking
Hualiang Xie
committed
Merge remote-tracking branch 'origin/main' into hualxie/infer_automodel
Hualiang Xie
committed
refactor(perf): address PR review comments
Hualiang Xie
committed
refactor(perf): unify GenerationBenchmarkResult and display across runtimes
Hualiang Xie
committed
fix: add --model-id to ONNX tests for shared resolve_model_path
Hualiang Xie
committed
fix: cast onnx_for_trace to str for Path() mypy arg-type
Hualiang Xie
committed
fix: make --model-id optional for perf with ONNX files
Hualiang Xie
committed
feat(perf): align composite generation with genai defaults and logging
Hualiang Xie
committed

Commits on Jul 7, 2026

fix: restore 'directory' assertion in test_onnx_file_rejected
Hualiang Xie
committed
refactor(perf): remove console progress log from generation loop
Hualiang Xie
committed
feat(perf): add per-iteration timing log in generation loop
Hualiang Xie
committed
feat(perf): auto-infer task for composite models from HF config
Hualiang Xie
committed