[vlm] adjust max length for special tokens (#37342)

* update

* apply suggestion

* fix tests for main branch

* remove unused logger

* add special tokens in tests

* nit

* fix more tests

* fix test

* pg also
This commit is contained in:
Raushan Turganbay
2025-04-16 20:49:20 +02:00
committed by GitHub
parent c94c59fc47
commit 32eca7197a
39 changed files with 414 additions and 98 deletions

View File

@@ -360,3 +360,29 @@ class MllamaProcessorTest(ProcessorTesterMixin, unittest.TestCase):
len(inputs[self.text_input_name][0]) == len(inputs[self.text_input_name][1])
and len(inputs[self.text_input_name][1]) < 76
)
def test_special_mm_token_truncation(self):
"""Tests that special vision tokens do not get truncated when `truncation=True` is set."""
processor = self.get_processor()
input_str = self.prepare_text_inputs(batch_size=2, modality="image")
image_input = self.prepare_image_inputs(batch_size=2)
image_input = [[image_input[0]], [image_input[1]]]
_ = processor(
text=input_str,
images=image_input,
return_tensors="pt",
truncation=None,
padding=True,
)
with self.assertRaises(ValueError):
_ = processor(
text=input_str,
images=image_input,
return_tensors="pt",
truncation=True,
padding=True,
max_length=3,
)