Compare three model trainings with exact same seeding but different parallelization. (Losses must be 100% equivalent) #427
Description
Ensure that runs are deterministic even across different parallelization strategies.
Loss / Parameters / Outputs must be the same.