The matched-compute result is based on an effective-architecture theoretical FLOP estimator whose reported forward coefficients differ by model and whose training estimate multiplies forward computation by three to approximate forward plus backward work.

The matched-compute result is based on an effective-architecture theoretical FLOP estimator whose reported forward coefficients differ by model and whose training estimate multiplies forward computation by three to approximate forward plus backward work. · CiteArk

测量结果

已评估 0/8 项

获得支持 0有争议或冲突 0尚未评估 8

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
model: 47B classic · qualifier: approximately	3.8875 billion FLOPs per token	—	尚未评估
model: 63B classic · qualifier: approximately	5.1485 billion FLOPs per token	—	尚未评估
model: source · qualifier: approximately	3.0873 billion FLOPs per token	—	尚未评估
model: 67B SST · qualifier: approximately	5.4107 billion FLOPs per token	—	尚未评估
qualifier: approximation	3 factor	—	尚未评估
	4,096 tokens	—	尚未评估
	512 tokens	—	尚未评估
	128,896 rows	—	尚未评估

The matched-compute result is based on an effective-architecture theoretical FLOP estimator whose reported forward coefficients differ by model and whose training estimate multiplies forward computation by three to approximate forward plus backward work.

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果