feat(deepdata): 引入修改密码、文件预览组件与清洗算子增强，deepflow 改名为 deep-flow
本次同步 09-12 至 09-28 的代码变更，主要内容： 新增功能 - 修改密码：后端新增 ChangePasswordRequest DTO 与UserController/UserService 对应接口，前端新增 ChangePasswordDialog 并在 Header 接入入口 - 文件预览组件重构：file-preview 目录拆分出 VideoPreview（支持 blobUrl 直传）、 ImagePreview、AudioPreview，index.tsx 统一调度各类型预览器 - 新增 my_ops 算子模块：markitdown_parser、mineru_pdf_parser、text_quality_evaluator - 新增 tools/minio-file-server MinIO 文件服务 功能增强 - 数据清洗：cleaning_task_service/scheduler/routes 重构， data_juicer_executor 与 wrapper 增强，OperatorTable/LogsTable/TaskDetail 同步调整 - 数据标注：annotation sync.py、project.py 接口与schema 更新 - 数据采集、算子市场、首页与菜单路由相关页面调整 - DataJuicer 算子接入（useOperatorOperations、dj_ops_schema） 结构调整 - deepflow/ 更名为 deep-flow/（419 个文件重命名） - 新增 helm 部署编排、GitHub Actions 之外的自建镜像构建脚本 未纳入本次提交 - scripts-backup/、runtime/python-executor-backup/、components_backup/（备份目录） - *.backup*、__pycache__ 产物 - zh_core_web_sm-3.8.0-py3-none-any.whl（46MB 构建产物） - DataX/（内嵌独立仓库，远程亦未收录其源码）
This commit is contained in:
沃巴蒂
@@ -0,0 +1,229 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""
|
||||
Text Quality Evaluator Process Logic
|
||||
基于 LLM 的长文本多维度质量评估算子
|
||||
"""
|
||||
import re
|
||||
import time
|
||||
import json
|
||||
import logging
|
||||
import os
|
||||
import sys
|
||||
from typing import Dict, Any, List
|
||||
from concurrent.futures import ThreadPoolExecutor, as_completed
|
||||
|
||||
try:
|
||||
from deepdata.core.base_op import LLM
|
||||
from deepdata.common.utils.text_splitter import TextSplitter
|
||||
except ImportError:
|
||||
try:
|
||||
executor_path = os.path.abspath(os.path.join(os.path.dirname(__file__), '..', '..', 'runtime', 'python-executor'))
|
||||
if executor_path not in sys.path:
|
||||
sys.path.insert(0, executor_path)
|
||||
|
||||
from deepdata.core.base_op import LLM
|
||||
from deepdata.common.utils.text_splitter import TextSplitter
|
||||
except Exception as e:
|
||||
print(f"{e}")
|
||||
sys.exit(-1)
|
||||
|
||||
# 内置默认配置
|
||||
DEFAULT_DIMENSIONS = '[{"name": "连贯性", "desc": "逻辑通顺度"}, {"name": "信息密度", "desc": "有效信息占比"}]'
|
||||
DEFAULT_PROMPT = "请作为资深数据专家，对以下文本进行多维度质量打分（0.0~1.0）：\n\n【待评估文本】\n{context}\n\n【评估维度】\n{dimensions}\n\n请严格按照格式返回对应数量的逗号分隔的浮点数，不要输出其他内容。"
|
||||
|
||||
class TextQualityEvaluator(LLM):
|
||||
"""
|
||||
文本质量评估算子
|
||||
"""
|
||||
|
||||
def __init__(self, *args, **kwargs):
|
||||
# A. 解析 Headers (必须转为 Dict，否则 requests 报错)
|
||||
llm_headers = kwargs.get("LLMHeaders")
|
||||
if isinstance(llm_headers, str):
|
||||
try:
|
||||
llm_headers = json.loads(llm_headers)
|
||||
except:
|
||||
llm_headers = {"Content-Type": "application/json"}
|
||||
kwargs["LLMHeaders"] = llm_headers
|
||||
|
||||
# B. 解析 Body (必须转为 Dict)
|
||||
llm_body = kwargs.get("LLMBody")
|
||||
if isinstance(llm_body, str):
|
||||
try:
|
||||
llm_body = json.loads(llm_body)
|
||||
except:
|
||||
llm_body = {}
|
||||
kwargs["LLMBody"] = llm_body
|
||||
|
||||
# C. 转换 accessType (将 "true"/"false" 转为 布尔值)
|
||||
access_type = kwargs.get("accessType")
|
||||
if isinstance(access_type, str):
|
||||
kwargs["accessType"] = access_type.lower() == "true"
|
||||
|
||||
# D. 替换模型名称
|
||||
model_name = kwargs.get("LLMModel", "qwen-plus")
|
||||
if isinstance(llm_body, dict):
|
||||
llm_body["model"] = model_name
|
||||
kwargs["LLMBody"] = llm_body
|
||||
|
||||
# ==========================================
|
||||
# 3. 调用父类初始化
|
||||
# ==========================================
|
||||
super(TextQualityEvaluator, self).__init__(*args, **kwargs)
|
||||
|
||||
self.task_id = kwargs.get("taskId", "default_id")
|
||||
self.max_workers = int(kwargs.get("max_workers", 5))
|
||||
self.chunk_size = int(kwargs.get("chunk_size", 2000))
|
||||
self.chunk_overlap = int(kwargs.get("chunk_overlap", 100))
|
||||
|
||||
dim_conf = kwargs.get("dimensions")
|
||||
if not dim_conf: dim_conf = DEFAULT_DIMENSIONS
|
||||
if isinstance(dim_conf, str):
|
||||
try: self.dimensions = json.loads(dim_conf)
|
||||
except: self.dimensions = []
|
||||
else: self.dimensions = dim_conf
|
||||
|
||||
self.prompt_template = kwargs.get("prompt_template") or DEFAULT_PROMPT
|
||||
self.pattern = r'\d+\.\d+'
|
||||
|
||||
def _split_text(self, text: str) -> List[str]:
|
||||
"""文本分块逻辑"""
|
||||
if not text: return []
|
||||
|
||||
try:
|
||||
splitter = TextSplitter(1024 * 1024 * 10, self.chunk_size, self.chunk_overlap)
|
||||
return splitter.split_text(text)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 降级方案
|
||||
chunks = []
|
||||
start = 0
|
||||
while start < len(text):
|
||||
chunks.append(text[start : start + self.chunk_size])
|
||||
start += self.chunk_size - self.chunk_overlap
|
||||
return chunks
|
||||
|
||||
def execute(self, sample: Dict[str, Any]) -> Dict[str, Any]:
|
||||
"""算子执行入口"""
|
||||
start_time = time.time()
|
||||
|
||||
self.read_file_first(sample)
|
||||
|
||||
text = sample.get(getattr(self, 'text_key', 'text'), "")
|
||||
if not text: return sample
|
||||
|
||||
# 1. 分块
|
||||
chunks = self._split_text(text)
|
||||
print(f"📄 文本分块完成：共 {len(chunks)} 个块")
|
||||
|
||||
# 2. 准备 Prompt
|
||||
dim_str = "\n".join([f"- {d.get('name', 'Unknown')}: {d.get('desc', '')}" for d in self.dimensions])
|
||||
prompts = [self.prompt_template.replace("{context}", c).replace("{dimensions}", dim_str) for c in chunks]
|
||||
|
||||
# 3. 并发调用
|
||||
weighted_scores = [0.0] * len(self.dimensions)
|
||||
total_chars = 0
|
||||
|
||||
with ThreadPoolExecutor(max_workers=self.max_workers) as executor:
|
||||
future_to_idx = {executor.submit(self._call_llm_with_retry, p): i for i, p in enumerate(prompts)}
|
||||
for future in as_completed(future_to_idx):
|
||||
idx = future_to_idx[future]
|
||||
try:
|
||||
scores = future.result()
|
||||
if scores:
|
||||
for i, s in enumerate(scores):
|
||||
weighted_scores[i] += s * len(chunks[idx])
|
||||
total_chars += len(chunks[idx])
|
||||
except Exception as e:
|
||||
print(f"❌ 块 {idx} 失败: {e}")
|
||||
|
||||
# 4. 聚合
|
||||
final_scores = {}
|
||||
if total_chars > 0:
|
||||
for i, dim in enumerate(self.dimensions):
|
||||
final_scores[dim.get('name', f'Dim_{i}')] = round(weighted_scores[i] / total_chars, 4)
|
||||
|
||||
print(f"✅ 评估完成 | 耗时: {time.time()-start_time:.2f}s | 得分: {final_scores}")
|
||||
|
||||
output_record = {
|
||||
"evaluation_scores": final_scores,
|
||||
"dimensions_count": len(self.dimensions),
|
||||
# 添加关联元数据
|
||||
"source_file_id": sample.get(self.fileid_key), # 原始文件 ID
|
||||
"source_file_name": sample.get(self.filename_key), # 原始文件名
|
||||
"source_dataset_id": sample.get("dataset_id")
|
||||
}
|
||||
|
||||
sample[self.text_key] = json.dumps(output_record, ensure_ascii=False)
|
||||
sample["target_type"] = "jsonl"
|
||||
base_name = os.path.splitext(sample.get(self.filename_key, ""))[0]
|
||||
sample[self.filename_key] = f"{base_name}_eval"
|
||||
|
||||
'''
|
||||
# 独立存储文件
|
||||
try:
|
||||
self.save_sample([output_record], sample)
|
||||
sample_key = getattr(self, 'text_key', 'text')
|
||||
sample[sample_key] = ""
|
||||
except Exception as e:
|
||||
print(f"Failed to save sample: {e}")
|
||||
'''
|
||||
return sample
|
||||
|
||||
def _call_llm_with_retry(self, prompt: str, retries: int = 2) -> List[float]:
|
||||
retry_count = 0
|
||||
while retry_count <= retries:
|
||||
try:
|
||||
response = self.llm(prompt)
|
||||
return self._parse_scores(response)
|
||||
except Exception as e:
|
||||
print(f"⚠️ LLM 请求失败 ({retry_count+1}): {e}")
|
||||
retry_count += 1
|
||||
time.sleep(1)
|
||||
return None
|
||||
|
||||
def _parse_scores(self, text: str) -> List[float]:
|
||||
if not text: return None
|
||||
matches = re.findall(self.pattern, text)
|
||||
return [float(m) for m in matches if 0.0 <= float(m) <= 1.0] if matches else None
|
||||
|
||||
# ==========================================================
|
||||
# 🧪 本地测试入口 (实际调用 Ollama)
|
||||
# ==========================================================
|
||||
if __name__ == "__main__":
|
||||
print("🚀 启动 TextQualityEvaluator 测试...")
|
||||
|
||||
# 配置参数 (请确保 Ollama 正在运行且已下载模型)
|
||||
mock_kwargs = {
|
||||
# --- Ollama 连接配置 ---
|
||||
# 注意：如果你在 Docker 容器外运行，使用 localhost；如果在容器内运行，可能需要 host.docker.internal
|
||||
"LLMUrl": "http://ollama.ailab.ccoop.cc:18585/v1/chat/completions",
|
||||
|
||||
"accessType": "true", # 必须为 "true" 才能绕过 DeepData 内部证书校验
|
||||
"isHttps": "false",
|
||||
"LLMModel": "qwen2.5:7b-instruct-q4_0", # <--- 请修改为你 Ollama 中实际存在的模型
|
||||
|
||||
# 请求体模板
|
||||
"LLMBody": '{"model": "{model}", "messages": [{"role": "user", "content": "{prompt}"}]}',
|
||||
|
||||
# 业务参数
|
||||
"dimensions": '[{"name": "连贯性", "desc": "逻辑通顺度"}, {"name": "信息密度", "desc": "有效信息占比"}]',
|
||||
"max_workers": 1
|
||||
}
|
||||
|
||||
# 模拟输入样本
|
||||
sample_text = "DeepData 是一个强大的数据处理平台，支持多种文档格式的解析与处理。" * 20
|
||||
mock_sample = {"text": sample_text}
|
||||
|
||||
try:
|
||||
# 实例化算子 (这步会触发 LLM 初始化)
|
||||
evaluator = TextQualityEvaluator(**mock_kwargs)
|
||||
|
||||
# 执行
|
||||
evaluator.execute(mock_sample)
|
||||
|
||||
except Exception as e:
|
||||
print(f"\n❌ 测试失败: {e}")
|
||||
import traceback
|
||||
traceback.print_exc()
|
||||
Reference in New Issue
Block a user