auto-coder 0.1.374__py3-none-any.whl → 0.1.375__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.


This version of auto-coder might be problematic. Click here for more details.

@@ -0,0 +1,217 @@
1
+ import json
2
+ import os
3
+ import time
4
+ from typing import Any, Dict, Generator, List, Optional, Tuple, Union
5
+
6
+ import pathspec
7
+ from byzerllm import ByzerLLM
8
+ import byzerllm
9
+ from loguru import logger
10
+ import traceback
11
+
12
+ from autocoder.common import AutoCoderArgs, SourceCode
13
+ from importlib.metadata import version
14
+ from pydantic import BaseModel
15
+ from autocoder.common import openai_content as OpenAIContentProcessor
16
+ from autocoder.rag.long_context_rag import LongContextRAG
17
+ import json, os
18
+ from autocoder.agent.base_agentic.base_agent import BaseAgent
19
+ from autocoder.agent.base_agentic.types import AgentRequest
20
+ from autocoder.common import SourceCodeList
21
+ from autocoder.rag.tools import register_search_tool, register_recall_tool
22
+ from byzerllm.utils.types import SingleOutputMeta
23
+ from autocoder.utils.llms import get_single_llm
24
+ try:
25
+ from autocoder_pro.rag.llm_compute import LLMComputeEngine
26
+ pro_version = version("auto-coder-pro")
27
+ autocoder_version = version("auto-coder")
28
+ logger.warning(
29
+ f"auto-coder-pro({pro_version}) plugin is enabled in auto-coder.rag({autocoder_version})")
30
+ except ImportError:
31
+ logger.warning(
32
+ "Please install auto-coder-pro to enhance llm compute ability")
33
+ LLMComputeEngine = None
34
+
35
+
36
+ class RAGAgent(BaseAgent):
37
+ def __init__(self, name: str,
38
+ llm: Union[byzerllm.ByzerLLM, byzerllm.SimpleByzerLLM],
39
+ files: SourceCodeList,
40
+ args: AutoCoderArgs,
41
+ rag: LongContextRAG,
42
+ conversation_history: Optional[List[Dict[str, Any]]] = None):
43
+
44
+ self.default_llm = self.llm
45
+ self.context_prune_llm = self.default_llm
46
+ if self.default_llm.get_sub_client("context_prune_model"):
47
+ self.context_prune_llm = self.default_llm.get_sub_client("context_prune_model")
48
+
49
+ self.llm = self.default_llm
50
+ if self.default_llm.get_sub_client("agentic_model"):
51
+ self.llm = self.default_llm.get_sub_client("agentic_model")
52
+
53
+ self.rag = rag
54
+ super().__init__(name, self.llm, files, args, conversation_history, default_tools_list=["read_file"])
55
+ # 注册RAG工具
56
+ # register_search_tool()
57
+ register_recall_tool()
58
+
59
+ class AgenticRAG:
60
+ def __init__(
61
+ self,
62
+ llm: ByzerLLM,
63
+ args: AutoCoderArgs,
64
+ path: str,
65
+ tokenizer_path: Optional[str] = None,
66
+ ) -> None:
67
+ self.llm = llm
68
+ self.args = args
69
+ self.path = path
70
+ self.tokenizer_path = tokenizer_path
71
+ self.rag = LongContextRAG(llm=self.llm, args=self.args, path=self.path, tokenizer_path=self.tokenizer_path)
72
+
73
+
74
+ def build(self):
75
+ pass
76
+
77
+ def search(self, query: str) -> List[SourceCode]:
78
+ return []
79
+
80
+
81
+ def stream_chat_oai(
82
+ self,
83
+ conversations,
84
+ model: Optional[str] = None,
85
+ role_mapping=None,
86
+ llm_config: Dict[str, Any] = {},
87
+ extra_request_params: Dict[str, Any] = {}
88
+ ):
89
+ try:
90
+ return self._stream_chat_oai(
91
+ conversations,
92
+ model=model,
93
+ role_mapping=role_mapping,
94
+ llm_config=llm_config,
95
+ extra_request_params=extra_request_params
96
+ )
97
+ except Exception as e:
98
+ logger.error(f"Error in stream_chat_oai: {str(e)}")
99
+ traceback.print_exc()
100
+ return ["出现错误,请稍后再试。"], []
101
+
102
+ @byzerllm.prompt()
103
+ def conversation_to_query(self,messages: List[Dict[str, Any]]):
104
+ '''
105
+ 【历史对话】按时间顺序排列,从旧到新:
106
+ {% for message in messages %}
107
+ <message>
108
+ {% if message.role == "user" %}【用户】{% else %}【助手】{% endif %}
109
+ <content>
110
+ {{ message.content }}
111
+ </content>
112
+ </message>
113
+ {% endfor %}
114
+
115
+ 【当前问题】用户的最新需求如下:
116
+ <current_query>
117
+ {{ query }}
118
+ </current_query>
119
+ '''
120
+ temp_messages = messages[0:-1]
121
+ message = messages[-1]
122
+
123
+ return {
124
+ "messages": temp_messages,
125
+ "query":message["content"]
126
+ }
127
+
128
+
129
+ def system_prompt(self):
130
+ '''
131
+ 你是一个基于知识库的智能助手,我的核心能力是通过检索增强生成(RAG)技术来回答用户问题。
132
+
133
+ 你的工作流程如下:
134
+ 1. 当用户提出问题时,我会首先理解问题的核心意图和关键信息需求
135
+ 2. 你会从多个角度分析问题,确定最佳的检索策略和关键词,然后召回工具 recall 获取与问题最相关的详细内容,只有在特别有必要的情况下,你才回使用 read_file 来获得相关文件更详细的信息。
136
+ 5. 如果获得的信息足够回答用户问题,你会直接生成回答。
137
+ 6. 如果获得的信息不足以回答用户问题,你会继续使用 recall 工具,直到你确信已经获取了足够的信息来回答用户问题。
138
+ 7. 有的问题可能需要拆解成多个问题,分别进行recall,然后最终得到的结果才是完整信息,最后才能进行回答。
139
+
140
+ 此外,你回答会遵循以下要求:
141
+
142
+ 1. 严格基于召回的文档内容回答
143
+ - 如果召回的文档提供的信息无法回答问题,请明确回复:"抱歉,文档中没有足够的信息来回答这个问题。"
144
+ - 不要添加、推测或扩展文档未提及的信息
145
+
146
+ 2. 格式如 ![image](/path/to/images/path.png) 的 Markdown 图片处理
147
+ - 根据Markdown 图片前后文本内容推测改图片与问题的相关性,有相关性则在回答中输出该Markdown图片路径
148
+ - 根据相关图片在文档中的位置,自然融入答复内容,保持上下文连贯
149
+ - 完整保留原始图片路径,不省略任何部分
150
+
151
+ 3. 回答格式要求
152
+ - 使用markdown格式提升可读性
153
+ {% if local_image_host %}
154
+ 4. 图片路径处理
155
+ - 图片地址需返回绝对路径,
156
+ - 对于Windows风格的路径,需要转换为Linux风格, 例如:C:\\Users\\user\\Desktop\\image.png 转换为 C:/Users/user/Desktop/image.png
157
+ - 为请求图片资源 需增加 http://{{ local_image_host }}/static/ 作为前缀
158
+ 例如:/path/to/images/image.png, 返回 http://{{ local_image_host }}/static/path/to/images/image.png
159
+ {% endif %}
160
+ '''
161
+ return {
162
+ "local_image_host": self.args.local_image_host
163
+ }
164
+
165
+
166
+ def _stream_chat_oai(
167
+ self,
168
+ conversations,
169
+ model: Optional[str] = None,
170
+ role_mapping=None,
171
+ llm_config: Dict[str, Any] = {},
172
+ extra_request_params: Dict[str, Any] = {}
173
+ ):
174
+ if not llm_config:
175
+ llm_config = {}
176
+
177
+ if extra_request_params:
178
+ llm_config.update(extra_request_params)
179
+
180
+ conversations = OpenAIContentProcessor.process_conversations(conversations)
181
+
182
+ context = []
183
+
184
+ def _generate_sream():
185
+
186
+ recall_request = AgentRequest(user_input=self.conversation_to_query.prompt(conversations))
187
+ rag_agent = RAGAgent(
188
+ name="RAGAgent",
189
+ llm=self.llm,
190
+ files=SourceCodeList(sources=[]),
191
+ args=self.args,
192
+ rag=self.rag,
193
+ conversation_history=[]
194
+ )
195
+
196
+ rag_agent.who_am_i(self.system_prompt.prompt())
197
+
198
+ events =rag_agent.run_with_generator(recall_request)
199
+ for (t,content) in events:
200
+ if t == "thinking":
201
+ yield ("", SingleOutputMeta(
202
+ generated_tokens_count=0,
203
+ input_tokens_count=0,
204
+ reasoning_content=content,
205
+ ))
206
+ else:
207
+ yield (content, SingleOutputMeta(
208
+ generated_tokens_count=0,
209
+ input_tokens_count=0,
210
+ reasoning_content="",
211
+ ))
212
+
213
+ return _generate_sream(), context
214
+
215
+
216
+
217
+
@@ -0,0 +1,10 @@
1
+ # 导出 SearchTool 相关类和函数
2
+ from .search_tool import SearchTool, SearchToolResolver, register_search_tool
3
+
4
+ # 导出 RecallTool 相关类和函数
5
+ from .recall_tool import RecallTool, RecallToolResolver, register_recall_tool
6
+
7
+ __all__ = [
8
+ 'SearchTool', 'SearchToolResolver', 'register_search_tool',
9
+ 'RecallTool', 'RecallToolResolver', 'register_recall_tool'
10
+ ]
@@ -0,0 +1,162 @@
1
+ """
2
+ RecallTool 模块
3
+
4
+ 该模块实现了 RecallTool 和 RecallToolResolver 类,用于在 BaseAgent 框架中
5
+ 提供基于 LongContextRAG 的文档内容召回功能。
6
+ """
7
+
8
+ import os
9
+ import traceback
10
+ from typing import Dict, Any, List, Optional
11
+
12
+ from loguru import logger
13
+
14
+ from autocoder.agent.base_agentic.types import BaseTool, ToolResult
15
+ from autocoder.agent.base_agentic.tool_registry import ToolRegistry
16
+ from autocoder.agent.base_agentic.tools.base_tool_resolver import BaseToolResolver
17
+ from autocoder.agent.base_agentic.types import ToolDescription, ToolExample
18
+ from autocoder.common import AutoCoderArgs
19
+ from autocoder.rag.long_context_rag import LongContextRAG, RecallStat, ChunkStat, AnswerStat, RAGStat
20
+ from autocoder.rag.relevant_utils import FilterDoc, DocRelevance, DocFilterResult
21
+ from autocoder.common import SourceCode
22
+ from autocoder.rag.relevant_utils import TaskTiming
23
+
24
+
25
+ class RecallTool(BaseTool):
26
+ """召回工具,用于获取与查询相关的文档内容"""
27
+ query: str # 用户查询
28
+ file_paths: Optional[List[str]] = None # 指定要处理的文件路径列表,如果为空则自动搜索
29
+
30
+
31
+ class RecallToolResolver(BaseToolResolver):
32
+ """召回工具解析器,实现召回逻辑"""
33
+ def __init__(self, agent, tool, args):
34
+ super().__init__(agent, tool, args)
35
+ self.tool: RecallTool = tool
36
+
37
+ def resolve(self) -> ToolResult:
38
+ """实现召回工具的解析逻辑"""
39
+ try:
40
+ # 获取参数
41
+ query = self.tool.query
42
+ file_paths = self.tool.file_paths
43
+ rag:LongContextRAG = self.agent.rag
44
+ # 构建对话历史
45
+ conversations = [
46
+ {"role": "user", "content": query}
47
+ ]
48
+
49
+ # 创建 RAGStat 对象
50
+
51
+ rag_stat = RAGStat(
52
+ recall_stat=RecallStat(total_input_tokens=0, total_generated_tokens=0),
53
+ chunk_stat=ChunkStat(total_input_tokens=0, total_generated_tokens=0),
54
+ answer_stat=AnswerStat(total_input_tokens=0, total_generated_tokens=0)
55
+ )
56
+
57
+ # 如果提供了文件路径,则直接使用;否则,执行搜索
58
+ if file_paths:
59
+
60
+ # 创建 FilterDoc 对象
61
+ relevant_docs = []
62
+ for file_path in file_paths:
63
+ try:
64
+ with open(file_path, 'r', encoding='utf-8') as f:
65
+ content = f.read()
66
+
67
+ source_code = SourceCode(
68
+ module_name=file_path,
69
+ source_code=content
70
+ )
71
+
72
+ doc = FilterDoc(
73
+ source_code=source_code,
74
+ relevance=DocRelevance(is_relevant=True, relevant_score=5), # 默认相关性
75
+ task_timing=TaskTiming()
76
+ )
77
+ relevant_docs.append(doc)
78
+ except Exception as e:
79
+ logger.error(f"读取文件 {file_path} 失败: {str(e)}")
80
+ else:
81
+ # 调用文档检索处理
82
+ generator = rag._process_document_retrieval(conversations, query, rag_stat)
83
+
84
+ # 获取检索结果
85
+ relevant_docs = None
86
+ for item in generator:
87
+ if isinstance(item, dict) and "result" in item:
88
+ relevant_docs = item["result"]
89
+
90
+ if not relevant_docs:
91
+ return ToolResult(
92
+ success=False,
93
+ message="未找到相关文档",
94
+ content=[]
95
+ )
96
+
97
+ # 调用文档分块处理
98
+ relevant_docs = [doc.source_code for doc in relevant_docs]
99
+ doc_chunking_generator = rag._process_document_chunking(relevant_docs, conversations, rag_stat, 0)
100
+
101
+ # 获取分块结果
102
+ final_relevant_docs = None
103
+ for item in doc_chunking_generator:
104
+ if isinstance(item, dict) and "result" in item:
105
+ final_relevant_docs = item["result"]
106
+
107
+ if not final_relevant_docs:
108
+ return ToolResult(
109
+ success=False,
110
+ message="文档分块处理失败",
111
+ content=[]
112
+ )
113
+
114
+ # 格式化结果
115
+ doc_contents = []
116
+ for doc in final_relevant_docs:
117
+ doc_contents.append({
118
+ "path": doc.module_name,
119
+ "content": doc.source_code
120
+ })
121
+
122
+ return ToolResult(
123
+ success=True,
124
+ message=f"成功召回 {len(doc_contents)} 个相关文档片段",
125
+ content=doc_contents
126
+ )
127
+
128
+ except Exception as e:
129
+ import traceback
130
+ return ToolResult(
131
+ success=False,
132
+ message=f"召回工具执行失败: {str(e)}",
133
+ content=traceback.format_exc()
134
+ )
135
+
136
+
137
+ def register_recall_tool():
138
+ """注册召回工具"""
139
+ # 准备工具描述
140
+ description = ToolDescription(
141
+ description="召回与查询相关的文档内容",
142
+ parameters="query: 搜索查询\nfile_paths: 指定要处理的文件路径列表(可选)",
143
+ usage="用于根据查询获取相关文档的内容片段"
144
+ )
145
+
146
+ # 准备工具示例
147
+ example = ToolExample(
148
+ title="召回工具使用示例",
149
+ body="""<recall>
150
+ <query>如何实现文件监控功能</query>
151
+ </recall>"""
152
+ )
153
+
154
+ # 注册工具
155
+ ToolRegistry.register_tool(
156
+ tool_tag="recall", # XML标签名
157
+ tool_cls=RecallTool, # 工具类
158
+ resolver_cls=RecallToolResolver, # 解析器类
159
+ description=description, # 工具描述
160
+ example=example, # 工具示例
161
+ use_guideline="此工具用于根据用户查询召回相关文档内容,返回经过分块和重排序的文档片段。适用于需要深入了解特定功能实现细节的场景。" # 使用指南
162
+ )
@@ -0,0 +1,125 @@
1
+ """
2
+ SearchTool 模块
3
+
4
+ 该模块实现了 SearchTool 和 SearchToolResolver 类,用于在 BaseAgent 框架中
5
+ 提供基于 LongContextRAG 的文档搜索功能。
6
+ """
7
+
8
+ import os
9
+ from typing import Dict, Any, List, Optional
10
+
11
+ from loguru import logger
12
+
13
+ from autocoder.agent.base_agentic.types import BaseTool, ToolResult
14
+ from autocoder.agent.base_agentic.tool_registry import ToolRegistry
15
+ from autocoder.agent.base_agentic.tools.base_tool_resolver import BaseToolResolver
16
+ from autocoder.agent.base_agentic.types import ToolDescription, ToolExample
17
+ from autocoder.common import AutoCoderArgs
18
+ from autocoder.rag.long_context_rag import LongContextRAG, RecallStat, ChunkStat, AnswerStat, RAGStat
19
+ from autocoder.rag.relevant_utils import FilterDoc, DocRelevance, DocFilterResult
20
+
21
+
22
+ class SearchTool(BaseTool):
23
+ """搜索工具,用于获取与查询相关的文件列表"""
24
+ query: str # 用户查询
25
+ max_files: Optional[int] = 10 # 最大返回文件数量
26
+
27
+
28
+ class SearchToolResolver(BaseToolResolver):
29
+ """搜索工具解析器,实现搜索逻辑"""
30
+ def __init__(self, agent, tool, args):
31
+ super().__init__(agent, tool, args)
32
+ self.tool: SearchTool = tool
33
+
34
+ def resolve(self) -> ToolResult:
35
+ """实现搜索工具的解析逻辑"""
36
+ try:
37
+ # 获取参数
38
+ query = self.tool.query
39
+ max_files = self.tool.max_files
40
+ rag = self.agent.rag
41
+ # 构建对话历史
42
+ conversations = [
43
+ {"role": "user", "content": query}
44
+ ]
45
+
46
+ # 创建 RAGStat 对象
47
+ rag_stat = RAGStat(
48
+ recall_stat=RecallStat(total_input_tokens=0, total_generated_tokens=0),
49
+ chunk_stat=ChunkStat(total_input_tokens=0, total_generated_tokens=0),
50
+ answer_stat=AnswerStat(total_input_tokens=0, total_generated_tokens=0)
51
+ )
52
+
53
+ # 调用文档检索处理
54
+ generator = rag._process_document_retrieval(conversations, query, rag_stat)
55
+
56
+ # 获取最终结果
57
+ result = None
58
+ for item in generator:
59
+ if isinstance(item, dict) and "result" in item:
60
+ result = item["result"]
61
+
62
+ if not result:
63
+ return ToolResult(
64
+ success=False,
65
+ message="未找到相关文档",
66
+ content=[]
67
+ )
68
+
69
+ # 格式化结果
70
+ file_list = []
71
+ for doc in result:
72
+ file_list.append({
73
+ "path": doc.source_code.module_name,
74
+ "relevance": doc.relevance.relevant_score if doc.relevance else 0,
75
+ "is_relevant": doc.relevance.is_relevant if doc.relevance else False
76
+ })
77
+
78
+ # 按相关性排序
79
+ file_list.sort(key=lambda x: x["relevance"], reverse=True)
80
+
81
+ # 限制返回数量
82
+ file_list = file_list[:max_files]
83
+
84
+ return ToolResult(
85
+ success=True,
86
+ message=f"成功检索到 {len(file_list)} 个相关文件",
87
+ content=file_list
88
+ )
89
+
90
+ except Exception as e:
91
+ import traceback
92
+ return ToolResult(
93
+ success=False,
94
+ message=f"搜索工具执行失败: {str(e)}",
95
+ content=traceback.format_exc()
96
+ )
97
+
98
+
99
+ def register_search_tool():
100
+ """注册搜索工具"""
101
+ # 准备工具描述
102
+ description = ToolDescription(
103
+ description="搜索与查询相关的文件",
104
+ parameters="query: 搜索查询\nmax_files: 最大返回文件数量(可选,默认为10)",
105
+ usage="用于根据查询找到相关的代码文件"
106
+ )
107
+
108
+ # 准备工具示例
109
+ example = ToolExample(
110
+ title="搜索工具使用示例",
111
+ body="""<search>
112
+ <query>如何实现文件监控功能</query>
113
+ <max_files>5</max_files>
114
+ </search>"""
115
+ )
116
+
117
+ # 注册工具
118
+ ToolRegistry.register_tool(
119
+ tool_tag="search", # XML标签名
120
+ tool_cls=SearchTool, # 工具类
121
+ resolver_cls=SearchToolResolver, # 解析器类
122
+ description=description, # 工具描述
123
+ example=example, # 工具示例
124
+ use_guideline="此工具用于根据用户查询搜索相关代码文件,返回文件路径及其相关性分数。适用于需要快速找到与特定功能或概念相关的代码文件的场景。" # 使用指南
125
+ )
autocoder/version.py CHANGED
@@ -1,2 +1,2 @@
1
1
 
2
- __version__ = "0.1.374"
2
+ __version__ = "0.1.375"