The performance of zero-shot CLIP in Table 1 is not the same as it in original CLIP paper

[MUVGuan]

Hello, I have two questions.

Q1.As shown in Table 1, the performance of zero-shot CLIP (Resnet-50) on ImageNet is 58.18, but in Table 11 of original CLIP paper it is 59.6, why is it different?

Q2.Why the performance of linear probe CLIP is lower than zero-shot CLIP in Table 1 (55.87<58.18)? After linear probe, the performance should be higher, but why it drops?