SigLIP tokenizer not enforcing use_fast=True

[yxchng]

System Info

• transformers version: 4.38.2
• Platform: Linux-4.18.0-477.27.1.el8_8.x86_64-x86_64-with-glibc2.28
• Python version: 3.10.13
• Huggingface_hub version: 0.21.4
• Safetensors version: 0.4.2
• Accelerate version: not installed
• Accelerate config: not found
• PyTorch version (GPU?): 2.2.1+cu121 (True)
• Tensorflow version (GPU?): not installed (NA)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using GPU in script?:
• Using distributed or parallel set-up in script?:

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

from transformers import AutoTokenizer
t=AutoTokenizer.from_pretrained('google/siglip-so400m-patch14-384', use_fast=True)
assert t.is_fast, 'tokenizer is not fast'
print('Success')

Expected behavior

print 'Success' which indicates use_fast=True

[ArthurZucker]

Hey! Siglip does not have a fast tokenizer:

transformers/src/transformers/models/auto/tokenization_auto.py

Line 404 in 6226b37

("siglip", ("SiglipTokenizer" if is_sentencepiece_available() else None, None)),

[yxchng]

I don't quite understand the line you linked. Why this model is different from others and does not have fast tokenizer?

[ArthurZucker]

That is because of

transformers/src/transformers/models/siglip/tokenization_siglip.py

Line 284 in 84d406c

def canonicalize_text(self, text, *, keep_punctuation_exact_string=None):

basically the equivalent fast tokenizer is not implemented because it needed a bit more work. I'll see if I have time to add it but otherwise it's a good issue!

[yxchng]

Is this going to get merged anytime soon?

[ArthurZucker]

I reviewed it, just need @NielsRogge 's updates