metrics don't match openai results

[rom1504]

https://github.com/cat-state/clip_benchmark/blob/main/plot.png?raw=true

any ideas why ?

[rom1504]

openai reports 62.5% on text to image retrieval @ 5 on coco
here found 40%

[rom1504]

LAION-AI/CLIP_benchmark#7 also wrong there