All 256 analyzable deployed chat tokenizers were forgeable through the plain encoder; 99.6% were also forgeable through template rendering, and enabling split_special_tokens still left 56.6% forgeable because 56.2% of models had at least one template-emitted control token not marked special.

All 256 analyzable deployed chat tokenizers were forgeable through the plain encoder; 99.6% were also forgeable through template rendering, and enabling split_special_tokens still left 56.6% forgeable because 56.2% of models had at least one template-emitted control token not marked special. · CiteArk

测量结果

已评估 0/38 项

获得支持 0有争议或冲突 0尚未评估 38

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
	400 repositories	—	尚未评估
	120 repositories	—	尚未评估
	24 repositories	—	尚未评估
	256 tokenizers	—	尚未评估
	99.6 percent of analyzed tokenizers	—	尚未评估
	56.2 percent of analyzed tokenizers	—	尚未评估
family: Qwen	59 tokenizers	—	尚未评估
family: Qwen · condition: default	100%	—	尚未评估
family: Qwen · condition: split_special_tokens enabled	84.7%	—	尚未评估
family: Nvidia	11 tokenizers	—	尚未评估
family: Nvidia · condition: default	100%	—	尚未评估
family: Nvidia · condition: split_special_tokens enabled	81.8%	—	尚未评估

All 256 analyzable deployed chat tokenizers were forgeable through the plain encoder; 99.6% were also forgeable through template rendering, and enabling split_special_tokens still left 56.6% forgeable because 56.2% of models had at least one template-emitted control token not marked special.

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果