perf(minimax-h3): reduce native runtime overhead - #1241

Draft

yifeif-nv wants to merge 8 commits into

NVIDIA:mainfrom

yifeif-nv:codex/minimax-h3-performance

Commits on Sep 10, 2026

feat(minimax-h3): unify native audiovisual delivery
yifeif-nv
committed
19e3759
perf(rtx): trim dynamic memory and plan load costs
yifeif-nv
committed
fff8121
test(minimax-h3): add resident video consumer
yifeif-nv
committed
4dcfdc7
perf(minimax-h3): size caches to each request
yifeif-nv
committed
aa594f6
docs(minimax-h3): document runtime reproduction
yifeif-nv
committed
c84a1a3
docs(minimax-h3): finalize measured delivery report
yifeif-nv
committed
f030c51
docs(minimax-h3): condense performance report
yifeif-nv
committed
348d0ee
docs(minimax-h3): lead with results and quick start
yifeif-nv
committed
2d8d27e