mirror of
https://github.com/NVIDIA/Model-Optimizer.git
synced 2026-10-30 07:34:44 +08:00
Three recipes shipped after the skill tree landed in #2201 and had no sheet: dflash2.yaml, lilicorr.yaml and lilicorr_conv.yaml. The stage procedures already applied to them; only the per-algorithm values were missing. dflash2.md and lilicorr.md follow the variant shape README.md defines -- delta only, pointing at dflash.md for the shared pipeline. One sheet covers both LiLiCorr recipes, since lilicorr_conv.yaml is the same algorithm with DFlash2's sublayer convolutions rather than a separate one; the delta has its own section. Facts are sourced from the recipes, the launcher examples and the plugin validation paths rather than from the papers, so the Known failures tables carry the error strings a run actually prints. Signed-off-by: Ye Yu <yeyu@nvidia.com> Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Ye Yu <yeyu@nvidia.com>