Why is the result of counting the same when the input text is different?

[wangyutian73]

Thank you for your excellent work! When I was testing, why is it that for different texts, such as "strawberries", "plates", they have different text_token, but the text_embedding output through CLIP is exactly the same? This results in different text input, but the results of the count are exactly the same.

[huiyan-dev]

Thank for your creative work. I reproduce the same result as @wangyutian73 said, would you please show me why and what can be gained from CLIP text prompt?