Create loader with test data.

Create logger for testing.

Create logger for testing.

Create logger for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Create checkpoint manager for testing.

Mock PIL Image for testing.

Mock lmdb module.

Create a mocked loader for testing.

Mock logger for testing.

Mock model for MLLM operators.

Create mock model.

Create mock model.

Create mock model.

OpenAI API response with reasoning_content.

Mock OpenAI API response.

Standard successful OpenAI API response.

Create pipeline with mocked sub-pipeline.

Create multi-stage operations config.

Create multi-turn DataItem for testing.

Create multi-turn DataItem.

Operator name for records.

Process a single item (delegates to process_batch).

Process by calling check() for each QA pair.

Process by calling rewrite() for each QA pair.

Check if any images have excluded extensions (item-level).

Normalize prompt formatting in questions (modifies question).

Remove redundant instruction patterns from multi-turn (item-level).

Normalize image tags - move all to first question (item-level).

Process a batch of items. Default implementation calls process() for each item. Override for batch optimization (e.g., MLLM

Process batch - returns keep result for each item.

Load and return the prompt text (lazy).

Configure pytest markers.

Number of QA pairs.

Get all QA pairs (cached).

Calculate QA-level reject rate.

Create a real PIL Image for testing.

Create clean registry for testing.

Create registry with test module.

Create a registry with a simple test class registered.

Create registry with registered modules.

Create registry with registered modules.

Calculate reject rate (item level).

Create request builder config.

Create request builder config.

Create request builder config.

Create request builder config.

Remove Reasoning: prefix from an answer.

Not used - this rewriter modifies questions.

Not used - item-level rewriter.

Not used - this rewriter operates at item level.

Normalize think tags in an answer.

Add empty think tags to a response missing them.

Calculate rewrite rate.

Create rewriter instance.

Create rewriter instance.

Create rewriter instance.

Create rewriter instance.

Create rewriter instance.

Create rewriter instance.

Create rewriter instance.

Create rewriter instance.

Create rewriter instance.

Create MLLMRewriter instance with rewrite_type='answer'.

Create MLLMRewriter instance.

Create MLLMRewriter instance with rewrite_type='question'.

Sample configuration dictionary.

Sample conversation data for testing.

Sample data item dictionary without image.

Sample data item with image.

List of sample data items.

Sample multi-turn conversation data.

Create a sample Result for testing.

Save data to JSON file. Args: data: List of data dictionaries. file_path: Output file path. indent: JSON

Save data to file. Args: data: List of data dictionaries. file_path: Output file path. auto_denormalize:

Save data to JSONL file. Args: data: List of data dictionaries. file_path: Output file path. ensure_ascii

Save configuration to YAML file. Args: config: StandardConfig to save. yaml_path: Output file path.

Create a saver instance for testing.

Create a saver instance for testing.

Create a saver instance for testing.

Shutdown and release resources. Workers are ephemeral (created per generate() call), nothing to clean up.

Create SimpleSaver instance.

Load and return the system prompt text (lazy).