Description #

Perplexity calculation operator – a text quality evaluation solution based on language models

Key features #

• Language model evaluation
• Calculates text perplexity based on the KenLM language model
• Supports quality evaluation for both Chinese and English texts
• Provides text readability metrics
• Quality evaluation
• The lower the perplexity, the higher the text quality
• Suitable for text quality filtering and evaluation

Technical implementation #

• Model core
• KenLM: Efficient language model inference
• SentencePiece: Chinese and English tokenization processing
• Computation optimization
• Batch processing to improve efficiency
• Memory-friendly model loading

Operator parameters #

Input #

Output #

Perplexity value column, with elements of floating-point type.

Parameters #

If a parameter does not have a default value, it is required.

Examples #

The following code demonstrates how to use daft to run the operator to calculate text perplexity for evaluating text quality.

from __future__ import annotations

import os

import daft
from daft import col
from daft.las.functions.text.perplexity_calculator import PerplexityCalculator
from daft.las.functions.udf import las_udf

if __name__ == "__main__":

    if os.getenv("DAFT_RUNNER", "native") == "ray":
        import logging

        import ray

        def configure_logging():
            logging.basicConfig(
                level=logging.INFO,
                format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
                datefmt="%Y-%m-%d %H:%M:%S.%s".format(),
            )
            logging.getLogger("tracing.span").setLevel(logging.WARNING)
            logging.getLogger("daft_io.stats").setLevel(logging.WARNING)
            logging.getLogger("DaftStatisticsManager").setLevel(logging.WARNING)
            logging.getLogger("DaftFlotillaScheduler").setLevel(logging.WARNING)
            logging.getLogger("DaftFlotillaDispatcher").setLevel(logging.WARNING)

        ray.init(dashboard_host="0.0.0.0", runtime_env={"worker_process_setup_hook": configure_logging})
        daft.set_runner_ray()
    daft.set_execution_config(actor_udf_ready_timeout=600)
    daft.set_execution_config(min_cpu_per_task=0)

    samples = {
        "text": [
            "",
            "",
            " 12345 !@#$%",
        ]
    }
    lang = "zh"

    ds = daft.from_pydict(samples)
    ds = ds.with_column(
        "perplexity",
        las_udf(
            PerplexityCalculator,
            construct_args={
                "lang": lang,
                "model_path": os.getenv("MODEL_PATH", "/opt/las/models"),
                "model_name": "kenlm/wikipedia",
            },
            num_gpus=0,
            batch_size=1,
            concurrency=1,
        )(col("text")),
    )

    ds.show()
    # ╭──────────────────────────────────────────────┬──────────────────────────────────────────────╮
    # │ text                                         ┆ perplexity                                  │
    # │ ---                                          ┆ ---                                          │
    # │ Utf8                                         ┆ Float64                                      │
    # ╞══════════════════════════════════════════════╪══════════════════════════════════════════════╡
    # │ Artificial intelligence technology is developing rapidly, and artificial intelligence technology has been widely app        ┆ 335.9                                        │
    # │ lied in various fields.                                 ┆                                             │
    # ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┤
    # │ This is a normal sentence.                            ┆ 530.3                                        │
    # ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┤
    # │ Garbled text 12345 !@#$%                           ┆ 9081.4                                       │
    # ╰──────────────────────────────────────────────┴──────────────────────────────────────────────╯