DPSKV3MFU Estimate MFU for DeepSeek V3 model. The MFU is around 39.0% for 4K context length pretraining phase.