List your company

Differential Transformer (Diff) implementation on New-feature tasks

38.2Session DurationPithTrain
37.74440.82243.946.978May 29, 2026

Evaluation results

MethodLinks
PithTrainFramework=PithTrain2026.0538.227.64769,50025,400
Megatron-LMFramework=Megatron-LM2026.0547.133.7125118,70057,100
TorchTitanFramework=TorchTitan2026.0549.640.358103,20036,000