opendcai/dataflow-webui/text2qa-sample-evaluator

Text2QASampleEvaluator Operator Reference

Text2QASampleEvaluator evaluates QA pairs across 4 dimensions, generating 8 output columns (grades + feedbacks for each dimension).

1. Import

from dataflow.operators.core_text import Text2QASampleEvaluator

2. Constructor

Text2QASampleEvaluator(
    llm_serving=llm_serving,
)

3. run() Signature

op.run(
    storage=self.storage.step(),
    input_question_key="question",
    input_answer_key="answer",
)
# returns: list of 8 output column names

4. Output Columns (8 columns)

Note: Column names use plural suffix (grades/feedbacks), not singular.

5. Usage Example

from dataflow.operators.core_text import Text2QASampleEvaluator
from dataflow.serving import APILLMServing_request
from dataflow.utils.storage import FileStorage
class MyPipeline:
    def __init__(self):
        self.storage = FileStorage(
            first_entry_file_name="./data/qa_pairs.jsonl",
            cache_path="./cache",
            file_name_prefix="step",
            cache_type="jsonl"
        )
        self.llm_serving = APILLMServing_request(
            api_url="https://api.openai.com/v1/chat/completions",
            key_name_of_api_key="DF_API_KEY",
            model_name="gpt-4o",
            max_workers=10
        )
        self.evaluator = Text2QASampleEvaluator(
            llm_serving=self.llm_serving
        )
    def forward(self):
        self.evaluator.run(
            storage=self.storage.step(),
            input_question_key="question",
            input_answer_key="answer"
        )
if __name__ == "__main__":
    pipeline = MyPipeline()
    pipeline.forward()

6. Runtime Logic

1. Read DataFrame from storage.
2. Validate input_question_key and input_answer_key columns exist.
3. Validate all 8 output columns do NOT exist (raises ValueError if they do).
4. For each row, call LLM 4 times (once per dimension):

• Question quality evaluation
• Answer alignment evaluation
• Answer verifiability evaluation
• Downstream value evaluation

1. Each LLM call returns a grade (numeric score) and feedback (text).
2. Write results to 8 output columns.
3. Return list of 8 output column names.

Text Constraints

• input_question_key and input_answer_key must contain non-empty text
• Empty or NaN values may cause evaluation errors
• No automatic text length limits enforced by operator
• LLM context window limits apply (typically 4K-128K tokens depending on model)

7. Important Notes

• Calls LLM 4 times per row (once per dimension), high cost
• All 8 output columns must not exist before calling
• No input_key parameter (will raise TypeError)