Create a temporary directory for tests.

Create a temporary JSON file with sample data.

Create a temporary JSONL file with sample data.

Create a temporary YAML config file.

Test large image count returns 392.

Test 10 images returns 672.

Test 11 images returns 448.

Test 15 images returns 448.

Test 16+ images returns 392.

Test 1 image returns the configured resize_image_size.

Test 2 images returns the configured resize_image_size.

Test 3 images returns 1024.

Test 5 images returns 1024.

Test 6 images returns 672.

Test that an absolute file_path is not modified.

Test _add_decision adds answer rewrite.

Test _add_decision adds keep decision (rejected=False).

Test _add_decision adds rejection.

Test adding file to generator.

Test add_filter for kept item.

Test adding multiple filter decisions.

Test add_filter_record adds record.

Test add_filter for rejection.

Test add_full_filter_record for full item rejection.

Test add_keep_record adds record.

Test adding rejected file.

Test add_rewrite.

Test adding multiple rewrite decisions.

Test add_rewrite_record adds record.

Test add_rewrite with new question.

Test batch size remains unchanged when adjustment disabled.

Test batch size adjustment when enabled.

Test when all items are rejected.

Test default allowed types.

Test applying multiple rewrites to multi-turn.

Test applying partial rejection to multi-turn.

Test apply_resize_image_pil.

Test apply_resize_image_pil with image larger than max_size.

Test apply_resize_image_pil with None input raises AttributeError.

Test apply_resize_image_pil with image already smaller than max_size.

Test apply_resize_image_pil with square image.

Test apply_to with filter keep.

Test apply_to with filter rejection.

Test apply_to with global rejection (qa_idx=-1).

Test apply_to with no decisions.

Test apply_to with rewrite.

Test getting attribute.

Test missing attribute returns None.

Test setting attribute.

Test that auto-completed progress dict has completed=True.

Test that BaseFormat cannot be instantiated directly.

Test that subclass must implement extensions property.

Test that subclass must implement load method.

Test that subclass must implement save method.

Test when batch size is larger than data size.

Test with batch size of 1.

Test rewrite with both new_question and new_answer None.

Test build_file creates directory.

Test build_file returns valid path.

Test build_file with nested directory.

Test build_from_cfg with missing 'type' key raises error.

Test build_from_cfg with None config returns None.

Test build_from_cfg passes extra kwargs to constructor.

Test build_from_cfg with unregistered type raises KeyError.

Test build_from_cfg with valid config and registry.

Test building from dict config.

Test building from string config (type name only).

Test building without type field raises error.

Test building with None config.

Test building nonexistent type raises error.

Test that loader options are passed correctly.

Test build_request creates request dict.

Test build_request with copy_img=True.

Test build_request includes image.

Test build_requests creates list of requests.

Test build_requests with batch_qa=True.

Test build_requests_selective.

Test building dataloaders with checkpoint support.

Test building with empty file list.

Test building with extra kwargs.

Test building dataloaders from file list.

Test building with invalid config raises error.

Test building with None dataset returns None.

Test building dataloaders from single file.

Test building dataloaders from YAML config.

Test building without logger.

Test building without work_dir disables checkpoint.

Test that filter operators are built correctly.

Test building multiple operators.

Test calculate_qa_statistics.

Test calculate_qa_statistics with empty data list.

Test calculate_qa_statistics with various batch sizes.

Test that __call__ accumulates data.

Test basic pipeline call.

Test basic pipeline execution.

Test pipeline call with empty data.

Test pipeline with empty data.

Test pipeline with multiple operators.