Skip to content

ccran / lufa-contract

Go to a project
Commit 6b4d3476 by ccran
Options

feat:添加merger合并同原文

parent 6b313617
Showing with 250 additions and 11 deletions
core/config.py
...@@ -16,7 +16,7 @@ MAX_SINGLE_CHUNK_SIZE=5000 ...@@ -16,7 +16,7 @@ MAX_SINGLE_CHUNK_SIZE=5000
META_KEY="META" META_KEY="META"
DEFAULT_RULESET_ID = "通用" DEFAULT_RULESET_ID = "通用"
ALL_RULESET_IDS = ["通用","借款","担保","财务口","金盘","金盘简化"] ALL_RULESET_IDS = ["通用","借款","担保","财务口","金盘","金盘简化"]
use_lufa = False use_lufa = True
if use_lufa: if use_lufa:
outer_backend_url = "http://znkf.lgfzgroup.com:48081" outer_backend_url = "http://znkf.lgfzgroup.com:48081"
base_fastgpt_url = "http://192.168.252.71:18089" base_fastgpt_url = "http://192.168.252.71:18089"
......
core/tools/memory_write.py
...@@ -28,6 +28,7 @@ class MemoryWriteTool(ToolBase): ...@@ -28,6 +28,7 @@ class MemoryWriteTool(ToolBase):
added: List[Dict] = [] added: List[Dict] = []
for f in findings: for f in findings:
finding_id = str(f.get("id") or "")
rule_title = f.get("rule_title") or "" rule_title = f.get("rule_title") or ""
issue = f.get("issue") or f.get("issue_description") or "" issue = f.get("issue") or f.get("issue_description") or ""
level = (f.get("level") or f.get("risk_level") or "M").upper() level = (f.get("level") or f.get("risk_level") or "M").upper()
...@@ -44,6 +45,7 @@ class MemoryWriteTool(ToolBase): ...@@ -44,6 +45,7 @@ class MemoryWriteTool(ToolBase):
issue=issue, issue=issue,
risk_level=level, risk_level=level,
suggestion=suggestion, suggestion=suggestion,
id=finding_id,
result=result, result=result,
) )
store.add_finding(finding_obj) store.add_finding(finding_obj)
......
core/tools/segment_merger.py 0 → 100644
from __future__ import annotations
import json
from typing import Any, Dict, List
from core.tool import tool, tool_func
from core.tools.segment_llm import LLMTool
from loguru import logger
MERGER_SYSTEM_PROMPT = '''
你是合同审查结果合并智能体（SegmentMerger）。
你的任务是：接收一组 findings，按 original_text 分组后合并。
【规则】
1. 以“文本重叠关系”分组：
- 完全相同：可合并。
- 存在公共子句（如一方是另一方子串，或两句共享连续核心片段）：可合并。
- 无明显公共子句、语义独立：不可合并。
2. 每个分组最终只保留 1 条 finding。
3. 对于“公共子句合并”的分组，合并后的 original_text 取“并集文本”：
- 若 A 是 B 的子串，取 B。
- 若 A、B 各有新增片段且围绕同一事实，拼接为不重复、语义通顺的一条完整原文。
4. 完全不相关的句子必须保留为不同分组。
【分组示例】
- 句子1："A：提交数据后15日内开票并付款。"
- 句子2："B：提交数据后15日内开票并付款。另由子公司承担服务费。"
- 句子3："甲方：xx公司"
应分为两组：
- 分组1（句子1+句子2）："提交数据后15日内开票并付款。另由子公司承担服务费。"
- 分组2（句子3）："甲方：xx公司"
【合并要求】
- 同步合并 issue 和 suggestion，兼顾组内要点
- 保留关键信息，不要机械拼接
【字段约束】
- 输出字段固定为：rule_title, segment_id, original_text, issue, risk_level, suggestion, result
- risk_level 仅允许 H/M/L/空字符串
- result 仅允许 合格/不合格/空字符串
- original_text 必须与该组合的原文一致
【输出要求】
- 严格输出 JSON
- 顶层结构必须是：{"findings": [...]}
'''
MERGER_USER_PROMPT = '''
【输入 findings】
{findings_json}
【任务】
请按“文本重叠关系（完全相同或存在公共子句）”分组后合并，并返回合并结果。
若同组文本可互补，请将 original_text 扩展为覆盖组内信息的并集文本；无关文本必须分到不同组。
输出 JSON。
'''
OUTPUT_EXAMPLE = '''
```json
{
"findings": [
{
"rule_title": "付款条款完整性",
"segment_id": 3,
"original_text": "甲方应于验收后30日内支付合同款项",
"issue": "付款期限明确，但未约定逾期违约责任，风险控制不足。",
"risk_level": "M",
"suggestion": "补充约定逾期付款违约金标准及计算方式。",
"result": "不合格"
}
]
}
```
'''
def _as_dict(item: Any) -> Dict[str, Any]:
if isinstance(item, dict):
return dict(item)
if hasattr(item, "__dict__"):
return dict(getattr(item, "__dict__"))
return {}
def _normalize_finding(raw: Dict[str, Any]) -> Dict[str, Any]:
risk_level = str(raw.get("risk_level", "") or "").upper()
if risk_level not in {"H", "M", "L", ""}:
risk_level = ""
result = str(raw.get("result", "") or "")
if result not in {"合格", "不合格", ""}:
result = ""
return {
"rule_title": str(raw.get("rule_title", "") or ""),
"segment_id": int(raw.get("segment_id", 0) or 0),
"original_text": str(raw.get("original_text", "") or ""),
"issue": str(raw.get("issue", "") or ""),
"risk_level": risk_level,
"suggestion": str(raw.get("suggestion", "") or ""),
"result": result,
}
@tool("segment_merger", "同证据 findings 合并")
class SegmentMergerTool(LLMTool):
def __init__(self) -> None:
super().__init__(MERGER_SYSTEM_PROMPT)
@tool_func(
{
"type": "object",
"properties": {
"findings": {"type": "array", "items": {"type": "object"}},
},
"required": ["findings"],
}
)
def run(self, findings: List[Dict[str, Any]]) -> Dict[str, List[Dict[str, Any]]]:
normalized_input = [_normalize_finding(_as_dict(item)) for item in (findings or [])]
if not normalized_input:
return {"findings": []}
msgs = self._build_prompt(normalized_input)
try:
resp = self.run_with_loop(self.chat_async(msgs))
data = self.parse_first_json(resp)
raw_findings = data.get("findings") or data.get("merged_findings") or []
if not isinstance(raw_findings, list):
raw_findings = []
normalized_output = [_normalize_finding(_as_dict(item)) for item in raw_findings]
return {"findings": normalized_output}
except Exception as e:
logger.error(f"SegmentMergerTool run error: {e}")
return {"findings": normalized_input}
def _build_prompt(self, findings: List[Dict[str, Any]]) -> List[Dict[str, str]]:
user_content = MERGER_USER_PROMPT.format(
findings_json=json.dumps(findings, ensure_ascii=False, indent=2)
) + OUTPUT_EXAMPLE
return self.build_messages(user_content)
if __name__ == "__main__":
tool = SegmentMergerTool()
sample = [
{
"rule_title": "支付时间审查",
"segment_id": 0,
"original_text": "本协议约定的服务内容全部履行完毕经甲方认可，在乙方提交检测数据后15个工作日内，乙方须向甲方提供相应数额的正规发票，甲方一次性支付合同总金额的100%给乙方。",
"issue": "付款条件缺乏实质把控。条款将付款绑定于提交数据，未明确'经甲方认可'的验收标准、期限及异议机制，且未设置质保金，存在验收流于形式即需全额付款的风险。",
"risk_level": "H",
"suggestion": "修改为：'乙方提交报告后，甲方在X个工作日内验收。验收合格且收到发票后15个工作日内支付95%；剩余5%作为质保金，满X个月无异议后无息支付。若验收不合格，甲方有权拒付并要求整改。'",
"result": "不合格"
},
{
"rule_title": "发票审查",
"segment_id": 0,
"original_text": "在乙方提交检测数据后15个工作日内，乙方须向甲方提供相应数额的正规发票，甲方一次性支付合同总金额的100%给乙方。甲方指定由其全资子公司长沙高新控股集团有限公司（简称高新控股）承担并支付本合同约定的检查服务费，乙方向高新控股开具相应金额的增值税专用发票。",
"issue": "缺失发票税率约定。条款明确了发票类型和开具时间，但未约定适用税率，违反审查规则，可能导致后续开票金额争议或税务合规风险。",
"risk_level": "H",
"suggestion": "补充税率约定。建议在'乙方向高新控股开具相应金额的增值税专用发票'后补充：'（税率：6%）'或根据实际业务类型补充具体税率数值。",
"result": "不合格"
},{
"rule_title": "主体审查",
"segment_id": 0,
"original_text": "委托方（甲方）： 湖南麓谷发展集团有限公司... 甲方指定由其全资子公司长沙高新控股集团有限公司（简称高新控股）承担并支付本合同约定的检查服务费... 签章处：甲方：长沙高新控股集团有限公司",
"issue": "签约主体不一致。首部甲方为'湖南麓谷发展集团有限公司'，但签章处及付款义务主体变更为'长沙高新控股集团有限公司'，且未明确授权委托或变更确认条款，存在主体混同及履约风险。",
"risk_level": "H",
"suggestion": "统一合同主体名称。若确由子公司履约，应将首部及正文甲方统一修改为'长沙高新控股集团有限公司'；若由母公司签约，应在签章处由母公司盖章，并补充'指定子公司代为履行付款义务'条款。",
"result": "不合格"
}]
print(json.dumps(tool.run(sample), ensure_ascii=False, indent=2))
data/batch/batch.py
...@@ -14,7 +14,7 @@ from utils.http_util import upload_file, fastgpt_openai_chat, download_file ...@@ -14,7 +14,7 @@ from utils.http_util import upload_file, fastgpt_openai_chat, download_file
SUFFIX='_麓发改进' SUFFIX='_麓发改进'
batch_input_dir_path = 'jp-input' batch_input_dir_path = 'jp-input'
batch_output_dir_path = 'jp-all' batch_output_dir_path = 'jp-output-lufa-simple'
batch_size = 5 batch_size = 5
# 麓发fastgpt接口 # 麓发fastgpt接口
# url = 'http://192.168.252.71:18089/api/v1/chat/completions' # url = 'http://192.168.252.71:18089/api/v1/chat/completions'
......
data/benchmark/eval.py
...@@ -121,7 +121,7 @@ def _parse_args() -> argparse.Namespace: ...@@ -121,7 +121,7 @@ def _parse_args() -> argparse.Namespace:
parser.add_argument( parser.add_argument(
"--datasets-dir", "--datasets-dir",
type=Path, type=Path,
default=base / "results" / "jp-all-merge-prompt", default=base / "results" / "jp-output-renji",
help="Directory containing Word files with annotations.", help="Directory containing Word files with annotations.",
) )
parser.add_argument( parser.add_argument(
...@@ -133,13 +133,13 @@ def _parse_args() -> argparse.Namespace: ...@@ -133,13 +133,13 @@ def _parse_args() -> argparse.Namespace:
parser.add_argument( parser.add_argument(
"--val-dir", "--val-dir",
type=Path, type=Path,
default=base / "results" / "jp-all-merge-prompt-extracted", default=base / "results" / "jp-output-renji-extracted",
help="Directory to store extracted xlsx files for comparison.", help="Directory to store extracted xlsx files for comparison.",
) )
parser.add_argument( parser.add_argument(
"--strip-suffixes", "--strip-suffixes",
nargs="*", nargs="*",
default=['_麓发改进'], default=['_麓发改进','_人机交互'],
help=( help=(
"Optional filename suffixes to strip from generated val xlsx stems before " "Optional filename suffixes to strip from generated val xlsx stems before "
"comparison, e.g. --strip-suffixes _v1 _审阅版" "comparison, e.g. --strip-suffixes _v1 _审阅版"
......
main.py
...@@ -20,6 +20,9 @@ from core.tools.segment_review import SegmentReviewTool ...@@ -20,6 +20,9 @@ from core.tools.segment_review import SegmentReviewTool
from core.tools.segment_rule_router import SegmentRuleRouterTool from core.tools.segment_rule_router import SegmentRuleRouterTool
from core.tools.retrieve_reference import RetrieveReferenceTool from core.tools.retrieve_reference import RetrieveReferenceTool
from core.tools.reflect_retry import ReflectRetryTool from core.tools.reflect_retry import ReflectRetryTool
from core.tools.segment_merger import SegmentMergerTool
from core.memory import Finding
from core.memory import FINDING_KEY_MERGE, FINDING_KEY_REFLECT, FINDING_KEY_REVIEW
app = FastAPI(title="合同审查智能体", version="0.1.0") app = FastAPI(title="合同审查智能体", version="0.1.0")
TMP_DIR = Path(__file__).resolve().parent / "tmp" TMP_DIR = Path(__file__).resolve().parent / "tmp"
...@@ -29,6 +32,7 @@ review_tool = SegmentReviewTool() ...@@ -29,6 +32,7 @@ review_tool = SegmentReviewTool()
rule_router_tool = SegmentRuleRouterTool() rule_router_tool = SegmentRuleRouterTool()
reference_tool = RetrieveReferenceTool() reference_tool = RetrieveReferenceTool()
reflect_tool = ReflectRetryTool() reflect_tool = ReflectRetryTool()
merger_tool = SegmentMergerTool()
@app.post("/sleep") @app.post("/sleep")
...@@ -214,7 +218,9 @@ def review_segment(payload: SegmentReviewRequest) -> SegmentReviewResponse: ...@@ -214,7 +218,9 @@ def review_segment(payload: SegmentReviewRequest) -> SegmentReviewResponse:
# Persist findings to memory store # Persist findings to memory store
for f in result.get("findings", []) or []: for f in result.get("findings", []) or []:
try: try:
store.add_finding_from_dict({ store.add_finding(
FINDING_KEY_REVIEW,
Finding.from_dict({
"rule_title": f.get("rule_title", ""), "rule_title": f.get("rule_title", ""),
"segment_id": segment_idx, "segment_id": segment_idx,
"original_text": f.get("original_text",''), "original_text": f.get("original_text",''),
...@@ -223,6 +229,7 @@ def review_segment(payload: SegmentReviewRequest) -> SegmentReviewResponse: ...@@ -223,6 +229,7 @@ def review_segment(payload: SegmentReviewRequest) -> SegmentReviewResponse:
"result": f.get("result", ""), "result": f.get("result", ""),
"suggestion": f.get("suggestion", ""), "suggestion": f.get("suggestion", ""),
}) })
)
except Exception as e: except Exception as e:
logger.error(e) logger.error(e)
continue continue
...@@ -278,6 +285,19 @@ class ReflectReviewResponse(BaseModel): ...@@ -278,6 +285,19 @@ class ReflectReviewResponse(BaseModel):
findings: List[Dict] findings: List[Dict]
class MergerRequest(BaseModel):
conversation_id: str
segment_id: int
finding_key: Optional[str] = FINDING_KEY_REVIEW
class MergerResponse(BaseModel):
conversation_id: str
segment_id: int
finding_key: str
merged_findings: List[Dict]
@app.post("/segments/review/reflect", response_model=ReflectReviewResponse) @app.post("/segments/review/reflect", response_model=ReflectReviewResponse)
def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse: def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse:
store = get_cached_memory(payload.conversation_id) store = get_cached_memory(payload.conversation_id)
...@@ -289,7 +309,7 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse: ...@@ -289,7 +309,7 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse:
summary_keywords = reference_tool.summary_keywords([rule]) summary_keywords = reference_tool.summary_keywords([rule])
context_summaries_facts = store.search_facts(summary_keywords) context_summaries_facts = store.search_facts(summary_keywords)
# 查找审查规则对应的 findings # 查找审查规则对应的 findings
findings = [f.__dict__ for f in store.search_findings("", rule_title=payload.rule_title)] findings = [f.__dict__ for f in store.search_findings(FINDING_KEY_REVIEW, "", rule_title=payload.rule_title)]
final_findings = reflect_tool.run( final_findings = reflect_tool.run(
party_role=payload.party_role, party_role=payload.party_role,
rule=rule, rule=rule,
...@@ -299,7 +319,9 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse: ...@@ -299,7 +319,9 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse:
for f in final_findings or []: for f in final_findings or []:
try: try:
store.add_final_finding_from_dict({ store.add_finding(
FINDING_KEY_REFLECT,
Finding.from_dict({
"rule_title": f.get("rule_title", ""), "rule_title": f.get("rule_title", ""),
"segment_id": f.get("segment_id", 0), "segment_id": f.get("segment_id", 0),
"original_text": f.get("original_text", ""), "original_text": f.get("original_text", ""),
...@@ -308,7 +330,7 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse: ...@@ -308,7 +330,7 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse:
"suggestion": f.get("suggestion", ""), "suggestion": f.get("suggestion", ""),
"result": f.get("result", "") "result": f.get("result", "")
}) })
# print(f'len(store) final_findings:{len(store.final_findings)}') )
except Exception: except Exception:
continue continue
return ReflectReviewResponse( return ReflectReviewResponse(
...@@ -317,6 +339,46 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse: ...@@ -317,6 +339,46 @@ def reflect_review(payload: ReflectReviewRequest) -> ReflectReviewResponse:
findings=final_findings or [], findings=final_findings or [],
) )
@app.post("/segments/review/merger", response_model=MergerResponse)
def merge_segment_findings(payload: MergerRequest) -> MergerResponse:
store = get_cached_memory(payload.conversation_id)
source_key = payload.finding_key or FINDING_KEY_REVIEW
target_segment_id = payload.segment_id - 1
segment_findings = store.get_findings_by_segment(source_key, target_segment_id)
unqualified_findings = [f for f in segment_findings if (f.result or "").strip() == "不合格"]
merged_result = merger_tool.run([f.__dict__ for f in unqualified_findings])
merged_findings = merged_result.get("findings", []) or []
for f in merged_findings:
try:
store.add_finding(
FINDING_KEY_MERGE,
Finding.from_dict(
{
"id": f.get("id", ""),
"rule_title": f.get("rule_title", ""),
"segment_id": target_segment_id,
"original_text": f.get("original_text", ""),
"issue": f.get("issue", ""),
"risk_level": (f.get("risk_level") or f.get("level") or "").upper(),
"suggestion": f.get("suggestion", ""),
"result": f.get("result", ""),
}
)
)
except Exception as e:
logger.error(e)
continue
return MergerResponse(
conversation_id=payload.conversation_id,
segment_id=target_segment_id,
finding_key=source_key,
merged_findings=merged_findings,
)
######################################################################################################################## ########################################################################################################################
class ConversationResponse(BaseModel): class ConversationResponse(BaseModel):
conversation_id: str conversation_id: str
...@@ -363,7 +425,7 @@ class MemoryExportRequest(BaseModel): ...@@ -363,7 +425,7 @@ class MemoryExportRequest(BaseModel):
conversation_id: str conversation_id: str
file_ext: str file_ext: str
file_name: Optional[str] = None file_name: Optional[str] = None
export_final: Optional[bool] = False finding_key: Optional[str] = FINDING_KEY_REVIEW
class MemoryExportResponse(BaseModel): class MemoryExportResponse(BaseModel):
...@@ -387,7 +449,7 @@ def export_memory(payload: MemoryExportRequest) -> MemoryExportResponse: ...@@ -387,7 +449,7 @@ def export_memory(payload: MemoryExportRequest) -> MemoryExportResponse:
raise HTTPException(status_code=500, detail=f"Export failed: {exc}") raise HTTPException(status_code=500, detail=f"Export failed: {exc}")
try: try:
doc_res = store.export_findings_to_doc_comments(doc_obj,export_final=payload.export_final or False) doc_res = store.export_findings_to_doc_comments(doc_obj, finding_key=payload.finding_key or FINDING_KEY_REVIEW)
except Exception as exc: except Exception as exc:
traceback.print_exc() traceback.print_exc()
raise HTTPException(status_code=500, detail=f"Export doc comments failed: {exc}") raise HTTPException(status_code=500, detail=f"Export doc comments failed: {exc}")
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or sign in to comment