List your company

Training Throughput on DeepSeek-V2-Lite workload

114,600Training Throughput (tokens/s)PithTrain
72,48083,41594,350105,285May 29, 2026

Evaluation results

MethodLinks
PithTrainHardware=1×8-H100, Parallelism=PP2,DP1,CP1,EP4, SeqLen=2048, Precision=BF162026.05114,600
Megatron-LMHardware=1×8-H100, Parallelism=PP2,DP1,CP1,EP4, SeqLen=2048, Precision=BF162026.05107,300
TorchTitanHardware=1×8-H100, Parallelism=PP2,DP1,CP1,EP4, SeqLen=2048, Precision=BF162026.0574,100