mirror of
https://github.com/huggingface/transformers.git
synced 2025-07-04 05:10:06 +06:00
a31fa218ad
* initial design * update all video processors * add tests * need to add qwen2-vl (not tested yet) * add qwen2-vl in auto map * fix copies * isort * resolve confilicts kinda * nit: * qwen2-vl is happy now * qwen2-5 happy * other models are happy * fix copies * fix tests * add docs * CI green now? * add more tests * even more changes + tests * doc builder fail * nit * Update src/transformers/models/auto/processing_auto.py Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * small update * imports correctly * dump, otherwise this is getting unmanagebale T-T * dump * update * another update * update * tests * move * modular * docs * test * another update * init * remove flakiness in tests * fixup * clean up and remove commented lines * docs * skip this one! * last fix after rebasing * run fixup * delete slow files * remove unnecessary tests + clean up a bit * small fixes * fix tests * more updates * docs * fix tests * update * style * fix qwen2-5-vl * fixup * fixup * unflatten batch when preparing * dump, come back soon * add docs and fix some tests * how to guard this with new dummies? * chat templates in qwen * address some comments * remove `Fast` suffix * fixup * oops should be imported from transforms * typo in requires dummies * new model added with video support * fixup once more * last fixup I hope * revert image processor name + comments * oh, this is why fetch test is failing * fix tests * fix more tests * fixup * add new models: internvl, smolvlm * update docs * imprt once * fix failing tests * do we need to guard it here again, why? * new model was added, update it * remove testcase from tester * fix tests * make style * not related CI fail, lets' just fix here * mark flaky for now, filas 15 out of 100 * style * maybe we can do this way? * don't download images in setup class --------- Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com>
270 lines
13 KiB
Python
270 lines
13 KiB
Python
# Copyright 2024 The HuggingFace Team. All rights reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
import shutil
|
|
import tempfile
|
|
import unittest
|
|
|
|
import numpy as np
|
|
import torch
|
|
|
|
from transformers.testing_utils import require_vision
|
|
from transformers.utils import is_vision_available
|
|
|
|
from ...test_processing_common import ProcessorTesterMixin
|
|
|
|
|
|
if is_vision_available():
|
|
from transformers import PixtralProcessor
|
|
|
|
|
|
@require_vision
|
|
class PixtralProcessorTest(ProcessorTesterMixin, unittest.TestCase):
|
|
processor_class = PixtralProcessor
|
|
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
cls.url_0 = "https://www.ilankelman.org/stopsigns/australia.jpg"
|
|
cls.image_0 = np.random.randint(255, size=(3, 876, 1300), dtype=np.uint8)
|
|
cls.url_1 = "http://images.cocodataset.org/val2017/000000039769.jpg"
|
|
cls.image_1 = np.random.randint(255, size=(3, 480, 640), dtype=np.uint8)
|
|
cls.image_2 = np.random.randint(255, size=(3, 1024, 1024), dtype=np.uint8)
|
|
|
|
def setUp(self):
|
|
self.tmpdirname = tempfile.mkdtemp()
|
|
processor = PixtralProcessor.from_pretrained("mistral-community/pixtral-12b")
|
|
processor.save_pretrained(self.tmpdirname)
|
|
|
|
def tearDown(self):
|
|
shutil.rmtree(self.tmpdirname)
|
|
|
|
def test_image_token_filling(self):
|
|
processor = self.processor_class.from_pretrained(self.tmpdirname)
|
|
# Important to check with non square image
|
|
image = torch.randint(0, 2, (3, 500, 316))
|
|
expected_image_tokens = 640
|
|
image_token_index = 10
|
|
|
|
messages = [
|
|
{
|
|
"role": "user",
|
|
"content": [
|
|
{"type": "image"},
|
|
{"type": "text", "text": "What is shown in this image?"},
|
|
],
|
|
},
|
|
]
|
|
inputs = processor(
|
|
text=[processor.apply_chat_template(messages)],
|
|
images=[image],
|
|
return_tensors="pt",
|
|
)
|
|
image_tokens = (inputs["input_ids"] == image_token_index).sum().item()
|
|
self.assertEqual(expected_image_tokens, image_tokens)
|
|
|
|
def test_processor_with_single_image(self):
|
|
processor = self.processor_class.from_pretrained(self.tmpdirname)
|
|
prompt_string = "USER: [IMG]\nWhat's the content of the image? ASSISTANT:"
|
|
|
|
# Make small for checking image token expansion
|
|
processor.image_processor.size = {"longest_edge": 30}
|
|
processor.image_processor.patch_size = {"height": 2, "width": 2}
|
|
|
|
# Test passing in an image
|
|
inputs_image = processor(text=prompt_string, images=self.image_0, return_tensors="pt")
|
|
self.assertIn("input_ids", inputs_image)
|
|
self.assertTrue(len(inputs_image["input_ids"]) == 1)
|
|
self.assertIsInstance(inputs_image["input_ids"], torch.Tensor)
|
|
self.assertIsInstance(inputs_image["pixel_values"], torch.Tensor)
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([1, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
input_ids = inputs_image["input_ids"]
|
|
self.assertEqual(
|
|
input_ids[0].tolist(),
|
|
# Equivalent to "USER: [IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END]\nWhat's the content of the image? ASSISTANT:"
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 4701, 1307, 1278, 3937, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
# Test passing in a url
|
|
inputs_url = processor(text=prompt_string, images=self.url_0, return_tensors="pt")
|
|
self.assertIn("input_ids", inputs_url)
|
|
self.assertTrue(len(inputs_url["input_ids"]) == 1)
|
|
self.assertIsInstance(inputs_url["input_ids"], torch.Tensor)
|
|
self.assertIsInstance(inputs_image["pixel_values"], torch.Tensor)
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([1, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
input_ids = inputs_url["input_ids"]
|
|
self.assertEqual(
|
|
input_ids[0].tolist(),
|
|
# Equivalent to "USER: [IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END]\nWhat's the content of the image? ASSISTANT:"
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 4701, 1307, 1278, 3937, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
# Test passing inputs as a single list
|
|
inputs_image = processor(text=prompt_string, images=[self.image_0], return_tensors="pt")
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([1, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
self.assertEqual(
|
|
inputs_image["input_ids"][0].tolist(),
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 4701, 1307, 1278, 3937, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
# Test as nested single list
|
|
inputs_image = processor(text=prompt_string, images=[[self.image_0]], return_tensors="pt")
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([1, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
self.assertEqual(
|
|
inputs_image["input_ids"][0].tolist(),
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 4701, 1307, 1278, 3937, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
def test_processor_with_multiple_images_single_list(self):
|
|
processor = self.processor_class.from_pretrained(self.tmpdirname)
|
|
prompt_string = "USER: [IMG][IMG]\nWhat's the difference between these two images? ASSISTANT:"
|
|
|
|
# Make small for checking image token expansion
|
|
processor.image_processor.size = {"longest_edge": 30}
|
|
processor.image_processor.patch_size = {"height": 2, "width": 2}
|
|
|
|
# Test passing in an image
|
|
inputs_image = processor(text=prompt_string, images=[self.image_0, self.image_1], return_tensors="pt")
|
|
self.assertIn("input_ids", inputs_image)
|
|
self.assertTrue(len(inputs_image["input_ids"]) == 1)
|
|
self.assertIsInstance(inputs_image["input_ids"], torch.Tensor)
|
|
self.assertIsInstance(inputs_image["pixel_values"], torch.Tensor)
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([2, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
input_ids = inputs_image["input_ids"]
|
|
self.assertEqual(
|
|
input_ids[0].tolist(),
|
|
# Equivalent to ["USER: [IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END][IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END]\nWhat's the difference between these two images? ASSISTANT:"]
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 6592, 2396, 2576, 2295, 8061, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
# Test passing in a url
|
|
inputs_url = processor(text=prompt_string, images=[self.url_0, self.url_1], return_tensors="pt")
|
|
self.assertIn("input_ids", inputs_url)
|
|
self.assertTrue(len(inputs_url["input_ids"]) == 1)
|
|
self.assertIsInstance(inputs_url["input_ids"], torch.Tensor)
|
|
self.assertIsInstance(inputs_image["pixel_values"], torch.Tensor)
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([2, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
input_ids = inputs_url["input_ids"]
|
|
self.assertEqual(
|
|
input_ids[0].tolist(),
|
|
# Equivalent to ["USER: [IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END][IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END]\nWhat's the difference between these two images? ASSISTANT:"]
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 6592, 2396, 2576, 2295, 8061, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
# Test passing in as a nested list
|
|
inputs_url = processor(text=prompt_string, images=[[self.image_0, self.image_1]], return_tensors="pt")
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([2, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
self.assertEqual(
|
|
inputs_url["input_ids"][0].tolist(),
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 6592, 2396, 2576, 2295, 8061, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
def test_processor_with_multiple_images_multiple_lists(self):
|
|
processor = self.processor_class.from_pretrained(self.tmpdirname)
|
|
prompt_string = [
|
|
"USER: [IMG][IMG]\nWhat's the difference between these two images? ASSISTANT:",
|
|
"USER: [IMG]\nWhat's the content of the image? ASSISTANT:",
|
|
]
|
|
processor.tokenizer.pad_token = "</s>"
|
|
image_inputs = [[self.image_0, self.image_1], [self.image_2]]
|
|
|
|
# Make small for checking image token expansion
|
|
processor.image_processor.size = {"longest_edge": 30}
|
|
processor.image_processor.patch_size = {"height": 2, "width": 2}
|
|
|
|
# Test passing in an image
|
|
inputs_image = processor(text=prompt_string, images=image_inputs, return_tensors="pt", padding=True)
|
|
self.assertIn("input_ids", inputs_image)
|
|
self.assertTrue(len(inputs_image["input_ids"]) == 2)
|
|
self.assertIsInstance(inputs_image["input_ids"], torch.Tensor)
|
|
self.assertIsInstance(inputs_image["pixel_values"], torch.Tensor)
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([3, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
input_ids = inputs_image["input_ids"]
|
|
self.assertEqual(
|
|
input_ids[0].tolist(),
|
|
# Equivalent to ["USER: [IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END][IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END]\nWhat's the difference between these two images? ASSISTANT:"]
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 6592, 2396, 2576, 2295, 8061, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
# Test passing in a url
|
|
inputs_url = processor(text=prompt_string, images=image_inputs, return_tensors="pt", padding=True)
|
|
self.assertIn("input_ids", inputs_url)
|
|
self.assertTrue(len(inputs_url["input_ids"]) == 2)
|
|
self.assertIsInstance(inputs_url["input_ids"], torch.Tensor)
|
|
self.assertIsInstance(inputs_image["pixel_values"], torch.Tensor)
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([3, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
input_ids = inputs_url["input_ids"]
|
|
self.assertEqual(
|
|
input_ids[0].tolist(),
|
|
# Equivalent to ["USER: [IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END][IMG][IMG][IMG_BREAK][IMG][IMG][IMG_END]\nWhat's the difference between these two images? ASSISTANT:"]
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 6592, 2396, 2576, 2295, 8061, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
# Test passing as a single flat list
|
|
inputs_image = processor(
|
|
text=prompt_string, images=[self.image_0, self.image_1, self.image_2], return_tensors="pt", padding=True
|
|
)
|
|
self.assertTrue(inputs_image["pixel_values"].shape == torch.Size([3, 3, 32, 32]))
|
|
|
|
# fmt: off
|
|
self.assertEqual(
|
|
inputs_image["input_ids"][0].tolist(),
|
|
[21510, 1058, 1032, 10, 10, 12, 10, 10, 13, 10, 10, 12, 10, 10, 13, 1010, 7493, 1681, 1278, 6592, 2396, 2576, 2295, 8061, 1063, 1349, 4290, 16002, 41150, 1058]
|
|
)
|
|
# fmt: on
|
|
|
|
def test_processor_returns_full_length_batches(self):
|
|
# to avoid https://github.com/huggingface/transformers/issues/34204
|
|
processor = self.processor_class.from_pretrained(self.tmpdirname)
|
|
prompt_string = [
|
|
"USER: [IMG]\nWhat's the content of the image? ASSISTANT:",
|
|
] * 5
|
|
processor.tokenizer.pad_token = "</s>"
|
|
image_inputs = [[self.image_0]] * 5
|
|
|
|
# Make small for checking image token expansion
|
|
processor.image_processor.size = {"longest_edge": 30}
|
|
processor.image_processor.patch_size = {"height": 2, "width": 2}
|
|
|
|
# Test passing in an image
|
|
inputs_image = processor(text=prompt_string, images=image_inputs, return_tensors="pt", padding=True)
|
|
self.assertIn("input_ids", inputs_image)
|
|
self.assertTrue(len(inputs_image["input_ids"]) == 5)
|
|
self.assertTrue(len(inputs_image["pixel_values"]) == 5)
|