Does params in table1 reported in the paper just contain the visual parameter number?
Hello, I found that the number of parameters of the visual component of clip-vit-L-14 is just 289.88 M. If we conside the actual size the parameters occupy, they occupy 289.88*2=579.76 M bytes( considering fp16), neither of which is consistent with the reported params.
关闭于 2024-10-17 1 条评论