feat: add party role check;

core/config.py

@@ -46,10 +46,11 @@ FILE_SUFFIX = "-审核批注"
 
 
 ## 关键参数**
 ## 关键参数**
 use_non_fastgpt_llm = False
 use_non_fastgpt_llm = False
-use_lufa = False
+use_lufa = True
 use_jp_machine = True
 use_jp_machine = True
 
 
 ## 关键参数**
 ## 关键参数**
+max_model_len = 131072
 ocr_url = 'http://192.168.252.71:8202/openapi/ocrUploadFile'
 ocr_url = 'http://192.168.252.71:8202/openapi/ocrUploadFile'
 if use_lufa:
 if use_lufa:
     outer_backend_url = "http://znkf.lgfzgroup.com:48081"
     outer_backend_url = "http://znkf.lgfzgroup.com:48081"

core/tools/party_role.py

+from __future__ import annotations
+
+from typing import Dict
+
+from loguru import logger
+
+from core.tool import tool, tool_func
+from core.tools.segment_llm import LLMTool
+
+
+PARTY_ROLE_SYSTEM_PROMPT = """
+你是一名专业合同审查律师。
+"""
+
+
+PARTY_ROLE_USER_PROMPT = """
+请分析指定公司在合同中的实际商业角色。
+
+注意：
+- 不要仅依据“甲方/乙方”字样判断
+- 必须结合实际权利义务关系判断
+
+重点考虑：
+- 谁提出需求
+- 谁支付费用
+- 谁接受服务或交付成果
+- 谁拥有验收、管理、监督权
+- 谁负责提供产品或服务
+
+输出要求：
+
+只输出一个 JSON：
+
+{{
+  "party_role": "demand_side | supplier_side | unclear",
+  "reason": ""
+}}
+
+其中：
+- demand_side：需求方、委托人、采购方、客户、甲方等
+- supplier_side：供应商、服务方、承包方、实施方、乙方等
+- unclear：无法明确判断
+
+reason 要用一句话简要说明核心依据。
+
+禁止输出任何Markdown、代码块或额外文本。
+
+待分析公司：
+{company_name}
+
+合同内容：
+{contract_text}
+"""
+
+
+VALID_PARTY_ROLES = {"demand_side", "supplier_side", "mixed", "unclear"}
+
+PARTY_ROLES_MAP={
+    "demand_side": "需求方、委托人、采购方、客户、甲方等",
+    "supplier_side": "供应商、服务方、承包方、实施方、乙方等",
+    "unclear": "",
+}
+
+
+@tool("party_role", "识别指定公司在合同中的实际商业角色")
+class PartyRoleTool(LLMTool):
+    def __init__(self) -> None:
+        super().__init__(PARTY_ROLE_SYSTEM_PROMPT)
+
+    @tool_func(
+        {
+            "type": "object",
+            "properties": {
+                "company_name": {"type": "string"},
+                "contract_text": {"type": "string"},
+            },
+            "required": ["company_name", "contract_text"],
+        }
+    )
+    def run(self, company_name: str, contract_text: str) -> Dict[str, str]:
+        company_name = (company_name or "").strip()
+        contract_text = (contract_text or "").strip()
+        if not company_name or not contract_text:
+            return {"party_role": "", "reason": ""}
+
+        user_content = PARTY_ROLE_USER_PROMPT.format(
+            company_name=company_name,
+            contract_text=contract_text,
+        )
+        try:
+            resp = self.run_with_loop(self.chat_async(self.build_messages(user_content)))
+            data = self.parse_first_json(resp)
+            party_role = str(data.get("party_role", "")).strip()
+            reason = str(data.get("reason", "")).strip()
+            if party_role in VALID_PARTY_ROLES:
+                return {"party_role": PARTY_ROLES_MAP.get(party_role, ""), "reason": reason}
+        except Exception as exc:
+            logger.error("Party role detection LLM failed: %s", exc)
+
+        return {"party_role": "", "reason": ""}
+
+
+if __name__ == "__main__":
+    tool = PartyRoleTool()
+    print(
+        tool.run(
+            company_name="麓发公司",
+            contract_text="麓发委托乙公司开发系统，甲公司支付费用并负责验收。",
+        )
+    )

main.py

@@ -19,6 +19,7 @@ from core.config import (
     pdf_support_formats,
     pdf_support_formats,
     MERGE_RULE_PROMPT,
     MERGE_RULE_PROMPT,
     use_lufa,
     use_lufa,
+    max_model_len
 )
 )
 from core.tools.segment_summary import SegmentSummaryTool
 from core.tools.segment_summary import SegmentSummaryTool
 from core.tools.segment_review import SegmentReviewTool
 from core.tools.segment_review import SegmentReviewTool
@@ -29,6 +30,7 @@ from core.tools.reflect_retry import ReflectRetryTool
 from core.tools.segment_merger import SegmentMergerTool
 from core.tools.segment_merger import SegmentMergerTool
 from core.tools.fact_merger import FactMergerTool
 from core.tools.fact_merger import FactMergerTool
 from core.tools.ruleset_router import RulesetRouterTool
 from core.tools.ruleset_router import RulesetRouterTool
+from core.tools.party_role import PartyRoleTool
 from core.memory import Finding
 from core.memory import Finding
 from core.memory import FINDING_KEY_MERGE, FINDING_KEY_REFLECT, FINDING_KEY_REVIEW
 from core.memory import FINDING_KEY_MERGE, FINDING_KEY_REFLECT, FINDING_KEY_REVIEW
 
 
@@ -44,6 +46,7 @@ reflect_tool = ReflectRetryTool()
 merger_tool = SegmentMergerTool()
 merger_tool = SegmentMergerTool()
 fact_merger_tool = FactMergerTool()
 fact_merger_tool = FactMergerTool()
 ruleset_router_tool = RulesetRouterTool()
 ruleset_router_tool = RulesetRouterTool()
+party_role_tool = PartyRoleTool()
 
 
 
 
 @app.post("/sleep")
 @app.post("/sleep")
@@ -84,6 +87,17 @@ class RulesetRouteResponse(BaseModel):
     reason: str = ""
     reason: str = ""
 
 
 
 
+class PartyRoleRequest(BaseModel):
+    company_name: str = Field(..., description="待分析公司名称")
+    contract_text: str = Field(..., description="合同全文或合同片段内容")
+
+
+class PartyRoleResponse(BaseModel):
+    company_name: str
+    party_role: str
+    reason: Optional[str] = ""
+
+
 @app.post("/rulesets/route", response_model=RulesetRouteResponse)
 @app.post("/rulesets/route", response_model=RulesetRouteResponse)
 def route_ruleset(payload: RulesetRouteRequest) -> RulesetRouteResponse:
 def route_ruleset(payload: RulesetRouteRequest) -> RulesetRouteResponse:
     question = (payload.question or "").strip()
     question = (payload.question or "").strip()
@@ -98,6 +112,26 @@ def route_ruleset(payload: RulesetRouteRequest) -> RulesetRouteResponse:
     )
     )
 
 
 
 
+@app.post("/contracts/party-role", response_model=PartyRoleResponse)
+def detect_party_role(payload: PartyRoleRequest) -> PartyRoleResponse:
+    company_name = (payload.company_name or "").strip()
+    contract_text = (payload.contract_text or "").strip()
+    if not company_name:
+        raise HTTPException(status_code=400, detail="company_name cannot be empty")
+    if not contract_text:
+        raise HTTPException(status_code=400, detail="contract_text cannot be empty")
+
+    result = party_role_tool.run(
+        company_name=company_name,
+        contract_text=contract_text[:max_model_len],  # TODO 合同文本过长时，优先保留包含公司名称的片段内容，后续优化
+    )
+    return PartyRoleResponse(
+        company_name=company_name,
+        party_role=result.get("party_role", ""),
+        reason=result.get("reason", ""),
+    )
+
+
 @app.post("/documents/parse", response_model=DocumentParseResponse)
 @app.post("/documents/parse", response_model=DocumentParseResponse)
 async def parse_document(payload: DocumentParseRequest) -> DocumentParseResponse:
 async def parse_document(payload: DocumentParseRequest) -> DocumentParseResponse:
     if not payload.urls:
     if not payload.urls: