📘 概述

DebertaV3SampleEvaluator 是一个基于 Nvidia Deberta V3 模型的质量分类器，用于评估输入文本的质量并返回相应的分类结果。该算子能够加载预训练的 Deberta V3 模型，对指定的文本列进行逐一样本评估，并将质量分数或类别标签添加为新的输出列。

__init__函数

def __init__(self, model_name, model_cache_dir='./dataflow_cache', device='cuda')

init参数说明

参数名	类型	默认值	说明
model_name	str	必需	Deberta V3 预训练模型的名称或路径。
model_cache_dir	str	'./dataflow_cache'	用于存储下载模型的本地缓存目录。
device	str	'cuda'	计算设备，例如 'cuda' 或 'cpu'。

Prompt模板说明

Prompt 模板名称	主要用途	适用场景	特点说明

run函数

def run(self, storage: DataFlowStorage, input_key: str, output_key: str='Debertav3Score')

参数

名称	类型	默认值	说明
storage	DataFlowStorage	必需	数据流存储实例，负责读取与写入数据。
input_key	str	必需	输入列名，该列包含待评估的文本数据。
output_key	str	'Debertav3Score'	输出列名，用于存储模型生成的质量评估结果。

🧠 示例用法

from dataflow.operators.text_pt.eval import DebertaV3SampleEvaluator
from dataflow.utils.storage import FileStorage

# 准备数据和存储
storage = FileStorage(first_entry_file_name="pt_input.jsonl")

# 初始化并运行算子
deberta_evaluator = DebertaV3SampleEvaluator(
    model_name="Nvidia/deberta-v3-large-local"
)
deberta_evaluator.run(
    storage.step(),
    input_key='raw_content',
    output_key='Debertav3Score'
)

🧾 默认输出格式（Output Format）

字段	类型	说明
[input_key]	str	输入的原始文本数据。
[output_key]	str	模型生成的质量评估类别。

示例输入：

{
  "raw_content": "AMICUS ANTHOLOGIES, PART ONE (1965-1972)\nFebruary 23, 2017 Alfred Eaker Leave a comment\nWith Dr. Terror's House of Horrors (1965, directed by Freddie Francis and written by Milton Subotsky) Amicus Productions (spearheaded by Subotsky and Max Rosenberg, who previously produced for Hammer and was a cousin to Doris Wishman) established itself as a vital competitor to Hammer Studios..."
}

示例输出：

{
  "raw_content": "AMICUS ANTHOLOGIES, PART ONE (1965-1972)\nFebruary 23, 2017 Alfred Eaker Leave a comment\nWith Dr. Terror's House of Horrors (1965, directed by Freddie Francis and written by Milton Subotsky) Amicus Productions (spearheaded by Subotsky and Max Rosenberg, who previously produced for Hammer and was a cousin to Doris Wishman) established itself as a vital competitor to Hammer Studios...",
  "Debertav3Score": "High"
}