Table 5 reports the 125M, 2.5B-token head-structure ablation. Multi-input/Multi-value (MIS/MVA) has the lowest reported perplexity among the listed patterns; MQA/MCS and MKA/MES are substantially worse despite equal total state size. Parameter-matched multi-head patterns fall between these choices.

Table 5 reports the 125M, 2.5B-token head-structure ablation. Multi-input/Multi-value (MIS/MVA) has the lowest reported perplexity among the listed patterns; MQA/MCS and MKA/MES are substantially worse despite equal total state size. Parameter-matched multi-head patterns fall between these choices. · CiteArk

测量结果

已评估 0/35 项

获得支持 0有争议或冲突 0尚未评估 35

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。
论文报告	最近可用实测	证据判断
24 heads	—	尚未评估
1 heads	—	尚未评估
1 heads	—	尚未评估
24 heads	—	尚未评估
24 layers	—	尚未评估
126.5 million parameters	—	尚未评估
11.66 perplexity	—	尚未评估
24 heads	—	尚未评估
1 heads	—	尚未评估
24 heads	—	尚未评估
1 heads	—	尚未评估
24 layers	—	尚未评估

Table 5 reports the 125M, 2.5B-token head-structure ablation. Multi-input/Multi-value (MIS/MVA) has the lowest reported perplexity among the listed patterns; MQA/MCS and MKA/MES are substantially worse despite equal total state size. Parameter-matched multi-head patterns fall between these choices.

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果