AbstractPhil's picture
anima_closeout v11: first live run of bf16_master_weights crashed at the scheduler β€” ds_config['bf16'] made DeepSpeed wrap the client Adam (FP16_UnfusedOptimizer), which torch ConstantLR rejects. Fork 812449a: client-side MasterWeightsAdam (fp32 masters, still a torch Adam; engine stays on the exp004-proven no-wrapping path). Proven in pure torch incl. the exact exp004 freeze repro: -3.0 bf16 gate frozen under plain Adam, moves under MasterWeightsAdam. CELL 0 asserts the new mechanism. Caches from the 28-min build are all banked β€” next launch goes straight to training.
30a7bb7 verified