The reported end-to-end TFLOPs/s uses 6 × sequence_length × parameter_count + 12 × layer_count × hidden_dimension × sequence_length². The paper notes that the attention term could be halved for causal masking, but deliberately does not halve it to remain consistent with Megatron-LM and prior reporting.

The reported end-to-end TFLOPs/s uses 6 × sequence_length × parameter_count + 12 × layer_count × hidden_dimension × sequence_length². The paper notes that the attention term could be halved for causal masking, but deliberately does not halve it to remain consistent with Megatron-LM and prior reporting. · CiteArk

测量结果

已评估 0/3 项

获得支持 0有争议或冲突 0尚未评估 3

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
term: sequence_length × parameter_count	6 multiplier	—	尚未评估
term: layer_count × hidden_dimension × sequence_length²	12 multiplier	—	尚未评估
status: discussed but not used	2 divisor	—	尚未评估

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果