max_length of Siglip2

[Yu-xm]

"When using the standalone GemmaTokenizerFast make sure to pass padding="max_length" and max_length=64 as that’s how the model was trained." Does Siglip2 support longer text input? If the max_length is set to 256 or 512, will text exceeding 64 be truncated?

[mitscha]

SigLIP 2 was trained with text length 64. The big_vision Gemma tokenizer implementation will pad/truncate to 64 if you set length=64. I'm not sure how other implementations behave (it seems you're referencing the HF transformers implementation). It's unclear how model quality will change if you set the length/max_length to a different value (and resize the positional embedding of the text encoder accordingly), since it was trained with 64.