first commit

tests/models/colqwen2/test_processing_colqwen2.py

@@ -0,0 +1,293 @@
+# Copyright 2025 The HuggingFace Inc. 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.
+"""Testing suite for the ColQwen2 processor."""
+
+import unittest
+
+import torch
+from parameterized import parameterized
+
+from transformers.models.colqwen2.processing_colqwen2 import ColQwen2Processor
+from transformers.testing_utils import get_tests_dir, require_torch, require_vision
+from transformers.utils import is_vision_available
+
+from ...test_processing_common import ProcessorTesterMixin
+
+
+if is_vision_available():
+    from transformers import (
+        ColQwen2Processor,
+    )
+
+SAMPLE_VOCAB = get_tests_dir("fixtures/test_sentencepiece.model")
+
+
+@require_torch
+@require_vision
+class ColQwen2ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
+    processor_class = ColQwen2Processor
+    model_id = "vidore/colqwen2-v1.0-hf"
+
+    @parameterized.expand([(1, "pt"), (2, "pt")])
+    @unittest.skip("Not tested before, to investigate")
+    def test_apply_chat_template_image(self, batch_size, return_tensors):
+        pass
+
+    @unittest.skip("ColQwen2Processor can only process one of text or images at a time")
+    def test_processor_with_multiple_inputs(self):
+        pass
+
+    @unittest.skip("ColQwen2Processor adds a prefix and suffix to the text")
+    def test_tokenizer_defaults(self):
+        pass
+
+    def test_get_num_vision_tokens(self):
+        "Tests general functionality of the helper used internally in vLLM"
+
+        processor = self.get_processor()
+
+        output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
+        self.assertTrue("num_image_tokens" in output)
+        self.assertEqual(len(output["num_image_tokens"]), 3)
+
+        self.assertTrue("num_image_patches" in output)
+        self.assertEqual(len(output["num_image_patches"]), 3)
+
+    def test_process_images(self):
+        # Processor configuration
+        image_input = self.prepare_image_inputs()
+        image_processor = self.get_component("image_processor")
+        tokenizer = self.get_component("tokenizer", max_length=112, padding="max_length")
+        image_processor.image_seq_length = 14
+
+        # Get the processor
+        processor = self.processor_class(
+            tokenizer=tokenizer,
+            image_processor=image_processor,
+        )
+
+        # Process the image
+        batch_feature = processor.process_images(images=image_input, return_tensors="pt")
+
+        # Assertions
+        self.assertIn("pixel_values", batch_feature)
+        self.assertEqual(batch_feature["pixel_values"].shape, torch.Size([1, 56, 1176]))
+
+    def test_process_queries(self):
+        # Inputs
+        queries = [
+            "Is attention really all you need?",
+            "Are Benjamin, Antoine, Merve, and Jo best friends?",
+        ]
+
+        # Processor configuration
+        image_processor = self.get_component("image_processor")
+        tokenizer = self.get_component("tokenizer", max_length=112, padding="max_length")
+        image_processor.image_seq_length = 14
+
+        # Get the processor
+        processor = self.processor_class(
+            tokenizer=tokenizer,
+            image_processor=image_processor,
+        )
+
+        # Process the image
+        batch_feature = processor.process_queries(text=queries, return_tensors="pt")
+
+        # Assertions
+        self.assertIn("input_ids", batch_feature)
+        self.assertIsInstance(batch_feature["input_ids"], torch.Tensor)
+        self.assertEqual(batch_feature["input_ids"].shape[0], len(queries))
+
+    # The following tests override the parent tests because ColQwen2Processor can only take one of images or text as input at a time.
+
+    def test_tokenizer_defaults_preserved_by_kwargs(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor_components["tokenizer"] = self.get_component("tokenizer", max_length=117, padding="max_length")
+
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+        input_str = self.prepare_text_inputs()
+        inputs = processor(text=input_str, return_tensors="pt")
+        self.assertEqual(inputs[self.text_input_name].shape[-1], 117)
+
+    def test_image_processor_defaults_preserved_by_image_kwargs(self):
+        """
+        We use do_rescale=True, rescale_factor=-1.0 to ensure that image_processor kwargs are preserved in the processor.
+        We then check that the mean of the pixel_values is less than or equal to 0 after processing.
+        Since the original pixel_values are in [0, 255], this is a good indicator that the rescale_factor is indeed applied.
+        """
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor_components["image_processor"] = self.get_component(
+            "image_processor", do_rescale=True, rescale_factor=-1.0
+        )
+        processor_components["tokenizer"] = self.get_component("tokenizer", max_length=117, padding="max_length")
+
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+
+        image_input = self.prepare_image_inputs()
+
+        inputs = processor(images=image_input, return_tensors="pt")
+        self.assertLessEqual(inputs[self.images_input_name][0][0].mean(), 0)
+
+    def test_kwargs_overrides_default_tokenizer_kwargs(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor_components["tokenizer"] = self.get_component("tokenizer", padding="longest")
+
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+        input_str = self.prepare_text_inputs()
+        inputs = processor(text=input_str, return_tensors="pt", max_length=112, padding="max_length")
+        self.assertEqual(inputs[self.text_input_name].shape[-1], 112)
+
+    def test_kwargs_overrides_default_image_processor_kwargs(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor_components["image_processor"] = self.get_component(
+            "image_processor", do_rescale=True, rescale_factor=1
+        )
+        processor_components["tokenizer"] = self.get_component("tokenizer", max_length=117, padding="max_length")
+
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+
+        image_input = self.prepare_image_inputs()
+
+        inputs = processor(images=image_input, do_rescale=True, rescale_factor=-1.0, return_tensors="pt")
+        self.assertLessEqual(inputs[self.images_input_name][0][0].mean(), 0)
+
+    def test_unstructured_kwargs(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+
+        input_str = self.prepare_text_inputs()
+        inputs = processor(
+            text=input_str,
+            return_tensors="pt",
+            do_rescale=True,
+            rescale_factor=-1.0,
+            padding="max_length",
+            max_length=76,
+        )
+
+        self.assertEqual(inputs[self.text_input_name].shape[-1], 76)
+
+    def test_unstructured_kwargs_batched(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+
+        image_input = self.prepare_image_inputs(batch_size=2)
+        inputs = processor(
+            images=image_input,
+            return_tensors="pt",
+            do_rescale=True,
+            rescale_factor=-1.0,
+            padding="longest",
+            max_length=76,
+        )
+
+        self.assertLessEqual(inputs[self.images_input_name][0][0].mean(), 0)
+
+    def test_doubly_passed_kwargs(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+
+        image_input = self.prepare_image_inputs()
+        with self.assertRaises(ValueError):
+            _ = processor(
+                images=image_input,
+                images_kwargs={"do_rescale": True, "rescale_factor": -1.0},
+                do_rescale=True,
+                return_tensors="pt",
+            )
+
+    def test_structured_kwargs_nested(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+
+        input_str = self.prepare_text_inputs()
+
+        # Define the kwargs for each modality
+        all_kwargs = {
+            "common_kwargs": {"return_tensors": "pt"},
+            "images_kwargs": {"do_rescale": True, "rescale_factor": -1.0},
+            "text_kwargs": {"padding": "max_length", "max_length": 76},
+        }
+
+        inputs = processor(text=input_str, **all_kwargs)
+        self.skip_processor_without_typed_kwargs(processor)
+
+        self.assertEqual(inputs[self.text_input_name].shape[-1], 76)
+
+    def test_structured_kwargs_nested_from_dict(self):
+        if "image_processor" not in self.processor_class.get_attributes():
+            self.skipTest(f"image_processor attribute not present in {self.processor_class}")
+        processor_components = self.prepare_components()
+        processor = self.processor_class(**processor_components)
+        self.skip_processor_without_typed_kwargs(processor)
+        image_input = self.prepare_image_inputs()
+
+        # Define the kwargs for each modality
+        all_kwargs = {
+            "common_kwargs": {"return_tensors": "pt"},
+            "images_kwargs": {"do_rescale": True, "rescale_factor": -1.0},
+            "text_kwargs": {"padding": "max_length", "max_length": 76},
+        }
+
+        inputs = processor(images=image_input, **all_kwargs)
+        self.assertEqual(inputs[self.text_input_name].shape[-1], 76)
+
+    # Can process only text or images at a time
+    def test_model_input_names(self):
+        processor = self.get_processor()
+        image_input = self.prepare_image_inputs()
+        inputs = processor(images=image_input)
+
+        self.assertSetEqual(set(inputs.keys()), set(processor.model_input_names))
+
+    @unittest.skip("ColQwen2Processor can't process text+image inputs at the same time")
+    def test_processor_text_has_no_visual(self):
+        pass
+
+    @unittest.skip("ColQwen2Processor adds a batch dimension to the pixel_values")
+    def test_image_processor_defaults(self):
+        pass
+
+    @unittest.skip("ColQwen2Processor can't process text+image inputs at the same time")
+    def test_get_num_multimodal_tokens_matches_processor_call(self):
+        pass
+
+    @unittest.skip("ColQwen2Processor can't process text+image inputs at the same time")
+    def test_flat_kwarg_applied_when_modality_dict_lacks_it(self):
+        pass