feat: add required fileds

app_gradio/app.py

@@ -247,32 +247,38 @@ def generate_llm_config_dataframe(llm_list):
 
 def suggest_fields_dataframe(rule_list, llm_list):
     """Suggest required field mappings based on selected evaluators"""
+    from dingo.io.input import RequiredField
+
     suggested_fields = set()
 
-    # Fields required by rule evaluators
-    rule_type_mapping = get_rule_type_mapping()
-    data_column_mapping = get_data_column_mapping()
+    # Get rule and llm name maps
+    rule_name_map = Model.get_rule_name_map()
+    llm_name_map = Model.get_llm_name_map()
 
+    # Fields required by rule evaluators
     for rule in rule_list:
-        # Find which type this rule belongs to
-        for rule_type, rules in rule_type_mapping.items():
-            if rule in rules:
-                if rule_type in data_column_mapping:
-                    suggested_fields.update(data_column_mapping[rule_type])
-                break
+        if rule in rule_name_map:
+            rule_class = rule_name_map[rule]
+            if hasattr(rule_class, '_required_fields'):
+                for field in rule_class._required_fields:
+                    if isinstance(field, RequiredField):
+                        suggested_fields.add(field.value)
 
     # Fields required by LLM evaluators
-    llm_column_mapping = get_llm_column_mapping()
     for llm in llm_list:
-        if llm in llm_column_mapping:
-            suggested_fields.update(llm_column_mapping[llm])
-
-    # Generate suggested fields rows
+        if llm in llm_name_map:
+            llm_class = llm_name_map[llm]
+            if hasattr(llm_class, '_required_fields'):
+                for field in llm_class._required_fields:
+                    if isinstance(field, RequiredField):
+                        suggested_fields.add(field.value)
+
+    # Generate suggested fields rows - Required Field and Dataset Column both with same value
     rows = []
     for field in sorted(suggested_fields):
         rows.append([field, field])
 
-    return gr.update(value=rows if rows else [["content", "content"]])
+    return gr.update(value=rows)
 
 
 def get_rule_type_mapping():
@@ -405,11 +411,11 @@ def get_data_column_mapping():
                     # Field mapping configuration
                     gr.Markdown("**EvalPipline.fields** - Field Mapping")
                     fields_dataframe = gr.Dataframe(
-                        value=[["content", "content"]],
-                        headers=["Field Key", "Dataset Column"],
+                        value=[],
+                        headers=["Required Field", "Dataset Column"],
                         datatype=["str", "str"],
                         column_count=(2, "fixed"),
-                        row_count=(1, "dynamic"),
+                        row_count=(0, "dynamic"),
                         label="Field Mappings (add/remove rows as needed)",
                         interactive=True
                     )

dingo/io/input/__init__.py

@@ -1 +1,2 @@
 from dingo.io.input.data import Data  # noqa E402.
+from dingo.io.input.required_field import RequiredField  # noqa E402.

dingo/io/input/required_field.py

@@ -0,0 +1,8 @@
+from enum import Enum
+
+
+class RequiredField(Enum):
+    CONTENT = "content"
+    PROMPT = "prompt"
+    CONTEXT = "context"
+    IMAGE = "image"

dingo/model/llm/agent/agent_hallucination.py

@@ -16,7 +16,7 @@
 import json
 from typing import Any, Dict, List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail, QualityLabel
 from dingo.model import Model
 from dingo.model.llm.agent.base_agent import BaseAgent
@@ -96,6 +96,8 @@ class AgentHallucination(BaseAgent):
 {{"claims": ["claim 1", "claim 2", ...]}}
 """
 
+    _required_fields = [RequiredField.CONTENT, RequiredField.CONTEXT]
+
     @classmethod
     def eval(cls, input_data: Data) -> EvalDetail:
         """

dingo/model/llm/agent/base_agent.py

@@ -13,7 +13,7 @@
 from abc import abstractmethod
 from typing import Any, Dict, List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail, QualityLabel
 from dingo.model.llm.agent.tools import ToolRegistry
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -48,6 +48,8 @@ class BaseAgent(BaseOpenAI):
     max_iterations: int = 5
     use_agent_executor: bool = False  # Opt-in to LangChain agent path
 
+    _required_fields = [RequiredField.CONTENT]
+
     @classmethod
     @abstractmethod
     def plan_execution(cls, input_data: Data) -> List[Dict[str, Any]]:

dingo/model/llm/agent/tools/base_tool.py

@@ -11,6 +11,8 @@
 
 from pydantic import BaseModel
 
+from dingo.io.input import RequiredField
+
 
 class ToolConfig(BaseModel):
     """Base configuration for tools"""
@@ -39,6 +41,8 @@ class BaseTool(ABC):
     description: str = None
     config: ToolConfig = ToolConfig()
 
+    _required_fields = [RequiredField.CONTENT]
+
     @classmethod
     @abstractmethod
     def execute(cls, **kwargs) -> Dict[str, Any]:

dingo/model/llm/agent/tools/tavily_search.py

@@ -21,6 +21,7 @@
 
 from pydantic import Field
 
+from dingo.io.input import RequiredField
 from dingo.model.llm.agent.tools.base_tool import BaseTool, ToolConfig
 from dingo.model.llm.agent.tools.tool_registry import tool_register
 from dingo.utils import log
@@ -76,6 +77,8 @@ class TavilySearch(BaseTool):
     description = "Search the web for factual information using Tavily AI"
     config: TavilyConfig = TavilyConfig()
 
+    _required_fields = [RequiredField.IMAGE]
+
     @classmethod
     def execute(cls, query: str, **kwargs) -> Dict[str, Any]:
         """

dingo/model/llm/base_lmdeploy_apiclient.py

@@ -5,7 +5,7 @@
 from pydantic import ValidationError
 
 from dingo.config.input_args import EvaluatorLLMArgs
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail, QualityLabel
 from dingo.model.llm.base import BaseLLM
 from dingo.model.response.response_class import ResponseScoreReason
@@ -15,6 +15,7 @@
 
 class BaseLmdeployApiClient(BaseLLM):
     dynamic_config = EvaluatorLLMArgs()
+    _required_fields = [RequiredField.CONTENT]  # Default, override in subclasses
 
     # @classmethod
     # def set_prompt(cls, prompt):

dingo/model/llm/base_openai.py

@@ -5,7 +5,7 @@
 from pydantic import ValidationError
 
 from dingo.config.input_args import EvaluatorLLMArgs
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail, QualityLabel
 from dingo.model.llm.base import BaseLLM
 from dingo.model.response.response_class import ResponseScoreReason
@@ -15,6 +15,7 @@
 
 class BaseOpenAI(BaseLLM):
     dynamic_config = EvaluatorLLMArgs()
+    _required_fields = [RequiredField.CONTENT]  # Default, override in subclasses
 
     # Embedding 模型配置（用于 RAG 相关评估器）
     embedding_model = None

dingo/model/llm/compare/llm_code_compare.py

@@ -2,7 +2,7 @@
 import re
 from typing import List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -25,6 +25,7 @@ class LLMCodeCompare(BaseOpenAI):
         'evaluation_results': ''
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
 你是一位专业的代码块识别评估专家,擅长分析 HTML 代码和 Markdown 文本中的代码块。现在我会提供三段内容：
 

dingo/model/llm/compare/llm_html_extract_compare.py

@@ -2,7 +2,7 @@
 import re
 from typing import List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -13,6 +13,7 @@
 
 @Model.llm_register("LLMHtmlExtractCompare")
 class LLMHtmlExtractCompare(BaseOpenAI):
+    _required_fields = [RequiredField.CONTENT]
     prompt = r"""
     你是一位专业的 HTML 内容提取评估专家，擅长分析 HTML 代码和 Markdown 文本的转换质量。现在我会提供三段内容：
 

dingo/model/llm/compare/llm_html_extract_compare_en.py

@@ -2,7 +2,7 @@
 import re
 from typing import List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -13,6 +13,7 @@
 
 @Model.llm_register("LLMHtmlExtractCompareEn")
 class LLMHtmlExtractCompareEn(BaseOpenAI):
+    _required_fields = [RequiredField.CONTENT]
     prompt = r"""
     You are a professional HTML content extraction evaluator, skilled in analyzing the conversion quality between HTML code and Markdown text. I will provide three pieces of content:
 

dingo/model/llm/compare/llm_html_extract_compare_v2.py

@@ -3,7 +3,7 @@
 
 import diff_match_patch as dmp_module
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -28,6 +28,7 @@ class LLMHtmlExtractCompareV2(BaseOpenAI):
     - input_data.raw_data.get("language", "en"): 语言类型 ("zh" 或 "en")
     """
 
+    _required_fields = [RequiredField.CONTENT, RequiredField.PROMPT]
     prompt = {
         "content_en": r"""Please compare the following two texts, each extracted from the same webpage using different HTML parsing methods. Your task is to determine whether there is a difference in the core informational content between them.
 

dingo/model/llm/compare/llm_math_compare.py

@@ -2,7 +2,7 @@
 import re
 from typing import List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -25,6 +25,7 @@ class LLMMathCompare(BaseOpenAI):
         'evaluation_results': ''
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
     你是一位专业的数学公式识别评估专家，擅长分析 HTML 代码和 Markdown 文本中的数学公式。现在我会提供三段内容：
 

dingo/model/llm/compare/llm_table_compare.py

@@ -2,7 +2,7 @@
 import re
 from typing import List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -25,6 +25,7 @@ class LLMTableCompare(BaseOpenAI):
         'evaluation_results': ''
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
     你是一位专业的表格识别评估专家,擅长分析 HTML 代码和 Markdown 文本中的表格。现在我会提供三段内容：
 

dingo/model/llm/hhh/llm_text_3h.py

@@ -1,5 +1,6 @@
 import json
 
+from dingo.io.input import RequiredField
 from dingo.io.output.eval_detail import EvalDetail, QualityLabel
 from dingo.model.llm.base_openai import BaseOpenAI
 from dingo.model.response.response_class import ResponseScoreReason
@@ -9,6 +10,8 @@
 
 # @Model.llm_register("LLMText3H")
 class LLMText3H(BaseOpenAI):
+    _required_fields = [RequiredField.CONTENT, RequiredField.PROMPT]
+
     @classmethod
     def build_messages(cls, input_data):
         question = input_data.prompt

dingo/model/llm/instruction_quality/llm_instruction_clarity.py

@@ -13,6 +13,7 @@
 4. Completeness: 指令是否完整，包含所有必要信息
 """
 
+from dingo.io.input import RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -42,6 +43,7 @@ class LLMInstructionClarity(BaseOpenAI):
         "examples": "examples/sft/evaluate_instruction_quality.py"
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
 # Role
 You are an expert in evaluating instruction quality for Large Language Model training data.

dingo/model/llm/instruction_quality/llm_task_difficulty.py

@@ -13,6 +13,7 @@
 4. Constraint Density: 约束条件密度
 """
 
+from dingo.io.input import RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -42,6 +43,7 @@ class LLMTaskDifficulty(BaseOpenAI):
         "examples": "examples/sft/evaluate_instruction_quality.py"
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
 # Role
 You are an expert in assessing task complexity and difficulty for LLM training data evaluation.

dingo/model/llm/llm_classify_qr.py

@@ -1,7 +1,7 @@
 import json
 from typing import List
 
-from dingo.io.input import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -20,6 +20,7 @@ class LLMClassifyQR(BaseOpenAI):
         "evaluation_results": ""
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
     'Classify the image into one of the following categories: "CAPTCHA", "QR code", or "Normal image". '
     'Return the type as the image category (CAPTCHA or QR code or Normal image) and the reason as the specific type of CAPTCHA or QR code. '

dingo/model/llm/llm_classify_topic.py

@@ -1,5 +1,6 @@
 import json
 
+from dingo.io.input import RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -22,6 +23,7 @@ class LLMClassifyTopic(BaseOpenAI):
         "validation_dataset": "AlignBench (https://github.com/THUDM/AlignBench)"
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
     Assume you are a topic classifier, and your task is to categorize user-provided instructions.
     There are six options in the list provided. You are required to select one category from the following list: ["Language Understanding and Processing", "Writing Ability", "Code", "Mathematics & Reasoning", "Task-oriented Role Play", "Knowledge-based Question and Answering"].

dingo/model/llm/llm_dataman_assessment.py

@@ -1,5 +1,6 @@
 import json
 
+from dingo.io.input import RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -25,6 +26,7 @@ class LLMDatamanAssessment(BaseOpenAI):
         "evaluation_results": ""
     }
 
+    _required_fields = [RequiredField.CONTENT]
     prompt = """
 ### Role
 You are an expert in data quality assessment for large language models.

dingo/model/llm/llm_document_parsing_ocr.py

@@ -2,7 +2,7 @@
 import re
 from typing import List
 
-from dingo.io import Data
+from dingo.io.input import Data, RequiredField
 from dingo.io.output.eval_detail import EvalDetail
 from dingo.model import Model
 from dingo.model.llm.base_openai import BaseOpenAI
@@ -20,6 +20,7 @@ class LLMMinerURecognizeQuality(BaseOpenAI):
         "description": "Evaluate the quality of mineru recognize",
         "evaluation_results": "error_category and error_label",
     }
+    _required_fields = [RequiredField.CONTENT, RequiredField.PROMPT]
     prompt = r"""
     你是一位熟悉文档解析领域的质量专家，你的核心任务是根据正确的markdown"工具标准结果Markdown"，以及对应OCR工具预测结果"Pred的内容"，获取工具预测结果的错误类型。
     *错误类别和标签*