diff --git a/.gitignore b/.gitignore index 9b3767bc..a381cf08 100644 --- a/.gitignore +++ b/.gitignore @@ -45,3 +45,4 @@ dark-server/dark-server.js node_modules/ gemini-websocket-proxy/ Tools/dark-server/dark-server.js +Tools/rimworld_cpt_data.jsonl diff --git a/Tools/Trainingdata_cpt.jsonl b/Tools/Trainingdata_cpt.jsonl new file mode 100644 index 00000000..2c4532b2 --- /dev/null +++ b/Tools/Trainingdata_cpt.jsonl @@ -0,0 +1,3 @@ +{"text":"基本竞争战略是由美国哈佛商学院著名的战略管理学家迈克尔·波特提出的,分别为:成本领先战略,差异化战略,集中化战略.企业必须从这三种战略中选择一种,作为其主导战略.要么把成本控制到比竞争者更低的程度;要么在企业产品和服务中形成与众不同的特色,让顾客感觉到你提供了比其他竞争者更多的价值;要么企业致力于服务于某一特定的市场细分,某一特定的产品种类或某一特定的地理范围."} +{"text":"交通运行监测调度中心,简称TOCC(Transportation Operations Coordination Center)TOCC围绕综合交通运输协调体系的构建,实施交通运行的监测,预测和预警,面向公众提供交通信息服务,开展多种运输方式的调度协调,提供交通行政管理和应急处置的信息保障.\nTOCC是综合交通运行监测协调体系的核心组成部分,实现了涵盖城市道路,高速公路,国省干线三大路网,轨道交通,地面公交,出租汽车三大市内交通方式,公路客运,铁路客运,民航客运三大城际交通方式的综合运行监测和协调联动,在综合交通的政府决策,行业监管,企业运营,百姓出行方面发挥了突出的作用."} +{"text":"美国职业摄影师协会(简称PPA)创立于1880年,是一个几乎与摄影术诞生历史一样悠久的享誉世界的非赢利性国际摄影组织,是由世界上54个国家的25000余名职业摄影师个人会员和近二百个附属组织和分支机构共同组成的,是世界上最大的专业摄影师协会.本世纪初PPA创立了美国视觉艺术家联盟及其所隶属的美国国际商业摄影师协会,美国新闻及体育摄影师协会,美国学生摄影联合会等组织.PPA在艺术,商业,纪实,体育等摄影领域一直引领世界潮流,走在世界摄影艺术与技术应用及商业规划管理的最前沿."} \ No newline at end of file diff --git a/Tools/convert_to_cpt.py b/Tools/convert_to_cpt.py new file mode 100644 index 00000000..c293f7bd --- /dev/null +++ b/Tools/convert_to_cpt.py @@ -0,0 +1,219 @@ +#!/usr/bin/env python3 +""" +RimWorld 代码转 CPT 训练数据脚本 +将 C# 源码和 XML Defs 转换为 JSONL 格式用于继续预训练 +支持智能分割超长文件 +""" + +import os +import json +import re +import argparse +from pathlib import Path +from typing import Generator, List + +# 排除的噪声文件（Unity 自动生成、二进制数据等） +EXCLUDED_FILES = { + 'UnitySourceGeneratedAssemblyMonoScriptTypes_v1.txt', + '__JobReflectionRegistrationOutput__', + 'AssemblyInfo.cs', + '.csproj', +} + +def should_exclude(file_path: Path) -> bool: + """检查文件是否应该被排除""" + name = file_path.name + for pattern in EXCLUDED_FILES: + if pattern in name: + return True + return False + +def read_file_content(file_path: Path) -> str: + """读取文件内容，处理多种编码""" + encodings = ['utf-8', 'utf-8-sig', 'gbk', 'latin-1'] + for encoding in encodings: + try: + with open(file_path, 'r', encoding=encoding) as f: + return f.read() + except UnicodeDecodeError: + continue + return None + +def split_csharp_content(content: str, max_length: int) -> List[str]: + """智能分割 C# 代码，尽量在类/方法边界分割""" + if len(content) <= max_length: + return [content] + + chunks = [] + # 按类定义分割 + class_pattern = r'(?=\n(?:public|private|internal|protected)?\s*(?:static\s+)?(?:partial\s+)?class\s+\w+)' + parts = re.split(class_pattern, content) + + current_chunk = "" + for part in parts: + if len(current_chunk) + len(part) <= max_length: + current_chunk += part + else: + if current_chunk: + chunks.append(current_chunk.strip()) + # 如果单个 part 仍然太长，按行分割 + if len(part) > max_length: + lines = part.split('\n') + current_chunk = "" + for line in lines: + if len(current_chunk) + len(line) + 1 <= max_length: + current_chunk += line + '\n' + else: + if current_chunk: + chunks.append(current_chunk.strip()) + current_chunk = line + '\n' + else: + current_chunk = part + + if current_chunk: + chunks.append(current_chunk.strip()) + + return [c for c in chunks if len(c) > 100] # 过滤太短的块 + +def split_xml_content(content: str, max_length: int) -> List[str]: + """智能分割 XML 内容，尽量在 Def 边界分割""" + if len(content) <= max_length: + return [content] + + chunks = [] + # 按顶级 Def 元素分割 + def_pattern = r'(?=\n\s*<[A-Z][a-zA-Z]*Def\s)' + parts = re.split(def_pattern, content) + + current_chunk = "" + for part in parts: + if len(current_chunk) + len(part) <= max_length: + current_chunk += part + else: + if current_chunk: + chunks.append(current_chunk.strip()) + if len(part) > max_length: + # 强制按长度分割 + for i in range(0, len(part), max_length - 200): + chunk = part[i:i + max_length - 200] + if len(chunk) > 100: + chunks.append(chunk.strip()) + current_chunk = "" + else: + current_chunk = part + + if current_chunk: + chunks.append(current_chunk.strip()) + + return [c for c in chunks if len(c) > 100] + +def create_csharp_entries(file_path: Path, content: str, max_length: int) -> List[dict]: + """创建 C# 代码的训练条目（支持分割）""" + relative_path = file_path.name + chunks = split_csharp_content(content, max_length - 50) # 留空间给 header + + entries = [] + for i, chunk in enumerate(chunks): + if len(chunks) > 1: + header = f"// File: {relative_path} (Part {i+1}/{len(chunks)})\n" + else: + header = f"// File: {relative_path}\n" + entries.append({"text": header + chunk}) + + return entries + +def create_xml_entries(file_path: Path, content: str, max_length: int) -> List[dict]: + """创建 XML 的训练条目（支持分割）""" + relative_path = file_path.name + chunks = split_xml_content(content, max_length - 50) + + entries = [] + for i, chunk in enumerate(chunks): + if len(chunks) > 1: + header = f"\n" + else: + header = f"\n" + entries.append({"text": header + chunk}) + + return entries + +def process_csharp_files(source_dir: Path, max_length: int) -> Generator[dict, None, None]: + """处理所有 C# 文件""" + for ext in ['*.cs', '*.txt']: + for file_path in source_dir.rglob(ext): + if file_path.stat().st_size < 100: + continue + if should_exclude(file_path): + continue + content = read_file_content(file_path) + if not content: + continue + if ext == '*.txt' and not ('class ' in content or 'namespace ' in content or 'public ' in content): + continue + for entry in create_csharp_entries(file_path, content, max_length): + yield entry + +def process_xml_files(source_dir: Path, max_length: int) -> Generator[dict, None, None]: + """处理所有 XML 文件""" + for file_path in source_dir.rglob('*.xml'): + if file_path.stat().st_size < 100: + continue + content = read_file_content(file_path) + if content: + for entry in create_xml_entries(file_path, content, max_length): + yield entry + +def main(): + parser = argparse.ArgumentParser(description='转换 RimWorld 代码为 CPT 训练数据') + parser.add_argument('--csharp-dir', type=str, help='C# 反编译代码目录') + parser.add_argument('--xml-dir', type=str, help='XML Defs 目录') + parser.add_argument('--output', type=str, default='rimworld_cpt_data.jsonl', help='输出文件路径') + parser.add_argument('--max-length', type=int, default=8000, help='单条数据最大字符数') + + args = parser.parse_args() + + entries = [] + stats = {'csharp': 0, 'xml': 0, 'split_chunks': 0} + + # 处理 C# 文件 + if args.csharp_dir: + csharp_path = Path(args.csharp_dir) + if csharp_path.exists(): + print(f"处理 C# 文件: {csharp_path}") + for entry in process_csharp_files(csharp_path, args.max_length): + entries.append(entry) + if "(Part " in entry['text'][:100]: + stats['split_chunks'] += 1 + else: + stats['csharp'] += 1 + + # 处理 XML 文件 + if args.xml_dir: + xml_path = Path(args.xml_dir) + if xml_path.exists(): + print(f"处理 XML 文件: {xml_path}") + for entry in process_xml_files(xml_path, args.max_length): + entries.append(entry) + if "(Part " in entry['text'][:100]: + stats['split_chunks'] += 1 + else: + stats['xml'] += 1 + + # 写入输出文件 + output_path = Path(args.output) + with open(output_path, 'w', encoding='utf-8') as f: + for entry in entries: + f.write(json.dumps(entry, ensure_ascii=False) + '\n') + + # 统计信息 + total_size = output_path.stat().st_size / (1024 * 1024) + print(f"\n=== 转换完成 ===") + print(f"C# 完整文件: {stats['csharp']}") + print(f"XML 完整文件: {stats['xml']}") + print(f"分割产生的块: {stats['split_chunks']}") + print(f"总条目数: {len(entries)}") + print(f"输出文件: {output_path}") + print(f"文件大小: {total_size:.2f} MB") + +if __name__ == '__main__': + main()