[GPU] Keep RMS normalization nodes in FP32 to prevent NaN from FP16 overflow#36504

Merged

mryzhov merged 14 commits into

openvinotoolkit:masterfrom

zhanmyz:cvs-187992-fix-ParlerTTSMini-issue-v3

Jul 9, 2026

Commits on Jun 22, 2026

[GPU] Keep RMS normalization nodes in FP32 to prevent NaN from FP16 overflow
zhanmyz
committed

Commits on Jun 23, 2026

[GPU] Add node existence checks to DisableFP16CompForRMSNormBlock unit test
zhanmyz
committed
[GPU] Add transformation_callback and match pipeline ConvertPrecision config
zhanmyz
committed
Potential fix for pull request finding

zhanmyz
and
Copilot
authored
[GPU] Add comprehensive test for RMS FP16 disable pass
zhanmyz
committed
[GPU] Rename pass to DisableFP16CompForAllRMS to match actual behavior (marks all RMS nodes).
zhanmyz
committed
[GPU] Rename RMS FP16 pass files to all-rms naming
zhanmyz
committed

Commits on Jun 24, 2026

Commits on Jun 25, 2026

[GPU] Refactor DisableFP16CompForAllRMS from ModelPass to MatcherPass
zhanmyz
committed

Commits on Jul 7, 2026

Commits on Jul 8, 2026

[Transformations] Use pattern predicate for MarkRMS instead of callback check
zhanmyz
committed