The attn_reg_transformer takes sentences separated by "< BRK >" as input, but I find that in scat's context data, there is no "< BRK >" to separate sentences. Does it affect the performance?
Meanwhile, for models which needs a separator between sentences, it might be difficult to use the scat data. Could you please provide a scat version with "< BRK >" separating sentences?
Dear author,
The attn_reg_transformer takes sentences separated by "< BRK >" as input, but I find that in scat's context data, there is no "< BRK >" to separate sentences. Does it affect the performance?
Meanwhile, for models which needs a separator between sentences, it might be difficult to use the scat data. Could you please provide a scat version with "< BRK >" separating sentences?