Fix： Fix dropout is null problem

repo_agent/chat_with_repo/gradio_interface.py

@@ -2,6 +2,7 @@
 import markdown
 from repo_agent.log import logger
 
+
 class GradioInterface:
     def __init__(self, respond_function):
         self.respond = respond_function
@@ -171,4 +172,4 @@ def respond_function(msg, system):
         """
         return msg, RAG, "Embedding_recall_output", "Key_words_output", "Code_output"
 
-    gradio_interface = GradioInterface(respond_function)
+    gradio_interface = GradioInterface(respond_function)

repo_agent/chat_with_repo/json_handler.py

@@ -2,25 +2,25 @@
 import sys
 from repo_agent.log import logger
 
+
 class JsonFileProcessor:
     def __init__(self, file_path):
         self.file_path = file_path
 
     def read_json_file(self):
-        # 读取 JSON 文件作为数据库
-        with open(self.file_path, 'r', encoding = 'utf-8') as file:
-            data = json.load(file)
-        return data
-    def extract_md_contents(self):
-        """
-        Extracts the contents of 'md_content' from a JSON file.
+        try:
+            with open(self.file_path, "r", encoding="utf-8") as file:
+                data = json.load(file)
+            return data
+        except FileNotFoundError:
+            logger.exception(f"File not found: {self.file_path}")
+            sys.exit(1)
 
-        Returns:
-            A list of strings representing the contents of 'md_content'.
-        """
+    def extract_data(self):
         # Load JSON data from a file
         json_data = self.read_json_file()
         md_contents = []
+        extracted_contents = []
         # Iterate through each file in the JSON data
         for file, items in json_data.items():
             # Check if the value is a list (new format)
@@ -31,64 +31,52 @@ def extract_md_contents(self):
                     if "md_content" in item and item["md_content"]:
                         # Append the first element of 'md_content' to the result list
                         md_contents.append(item["md_content"][0])
-        return md_contents
-
-    def extract_metadata(self):
-        """
-        Extracts metadata from JSON data.
-
-        Returns:
-            A list of dictionaries containing the extracted metadata.
-        """
-        # Load JSON data from a file
-        json_data = self.read_json_file()
-        extracted_contents = []
-        # Iterate through each file in the JSON data
-        for file_name, items in json_data.items():
-            # Check if the value is a list (new format)
-            if isinstance(items, list):
-                # Iterate through each item in the list
-                for item in items:
-                    # Build a dictionary containing the required information
-                    item_dict = {
-                        "type": item.get("type", "UnknownType"),
-                        "name": item.get("name", "Unnamed"),
-                        "code_start_line": item.get("code_start_line", -1),
-                        "code_end_line": item.get("code_end_line", -1),
-                        "have_return": item.get("have_return", False),
-                        "code_content": item.get("code_content", "NoContent"),
-                        "name_column": item.get("name_column", 0),
-                        "item_status": item.get("item_status", "UnknownStatus"),
-                        # Adapt or remove fields based on new structure requirements
-                    }
-                    extracted_contents.append(item_dict)
-            return extracted_contents
+                          # Build a dictionary containing the required information
+                        item_dict = {
+                            "type": item.get("type", "UnknownType"),
+                            "name": item.get("name", "Unnamed"),
+                            "code_start_line": item.get("code_start_line", -1),
+                            "code_end_line": item.get("code_end_line", -1),
+                            "have_return": item.get("have_return", False),
+                            "code_content": item.get("code_content", "NoContent"),
+                            "name_column": item.get("name_column", 0),
+                            "item_status": item.get("item_status", "UnknownStatus"),
+                            # Adapt or remove fields based on new structure requirements
+                        }
+                        extracted_contents.append(item_dict)
+        return md_contents,extracted_contents
 
-    def recursive_search(self, data_item, search_text, results):
+    def recursive_search(self, data_item, search_text, code_results, md_results):
         if isinstance(data_item, dict):
             # Direct comparison is removed as there's no direct key==search_text in the new format
             for key, value in data_item.items():
                 # Recursively search through dictionary values and lists
                 if isinstance(value, (dict, list)):
-                    self.recursive_search(value, search_text, results)
+                    self.recursive_search(value, search_text,code_results, md_results)
         elif isinstance(data_item, list):
             for item in data_item:
                 # Now we check for the 'name' key in each item of the list
                 if isinstance(item, dict) and item.get('name') == search_text:
                     # If 'code_content' exists, append it to results
                     if 'code_content' in item:
-                        results.append(item['code_content'])
+                        code_results.append(item['code_content'])
+                        md_results.append(item['md_content'])
                 # Recursive call in case of nested lists or dicts
-                self.recursive_search(item, search_text, results)
+                self.recursive_search(item, search_text, code_results, md_results)
 
     def search_code_contents_by_name(self, file_path, search_text):
         # Attempt to retrieve code from the JSON file
         try:
             with open(file_path, 'r', encoding='utf-8') as file:
                 data = json.load(file)
-                results = []  # List to store matching items' code_content
-                self.recursive_search(data, search_text, results)
-                return results if results else "No matching item found."
+                code_results = []
+                md_results = []  # List to store matching items' code_content and md_content
+                self.recursive_search(data, search_text, code_results, md_results)
+                # 确保无论结果如何都返回两个值
+                if code_results or md_results:
+                    return code_results, md_results
+                else:
+                    return ["No matching item found."], ["No matching item found."]
         except FileNotFoundError:
             return "File not found."
         except json.JSONDecodeError:
@@ -99,4 +87,4 @@ def search_code_contents_by_name(self, file_path, search_text):
 
 if __name__ == "__main__":
     processor = JsonFileProcessor("database.json")
-    md_contents = processor.extract_md_contents()
+    md_contents = processor.extract_md_contents()

repo_agent/chat_with_repo/main.py

@@ -9,12 +9,11 @@ def main():
     api_key = CONFIG["api_keys"][_model][0]["api_key"]
     api_base = CONFIG["api_keys"][_model][0]["base_url"]
     db_path = os.path.join(
-        CONFIG["repo_path"], CONFIG["project_hierarchy"], ".project_hierarchy.json"
+        CONFIG["repo_path"], CONFIG["project_hierarchy"], "project_hierarchy.json"
     )
 
     assistant = RepoAssistant(api_key, api_base, db_path)
-    md_contents = assistant.json_data.extract_md_contents()
-    meta_data = assistant.json_data.extract_metadata()
+    md_contents,meta_data = assistant.json_data.extract_data()
     assistant.chroma_data.create_vector_store(md_contents, meta_data)
     GradioInterface(assistant.respond)
 

repo_agent/chat_with_repo/prompt.py

@@ -2,7 +2,6 @@
 from repo_agent.log import logger
 from repo_agent.chat_with_repo.json_handler import JsonFileProcessor
 
-# logger.add("./log.txt", level="DEBUG", format="{time} - {name} - {level} - {message}")
 
 class TextAnalysisTool:
     def __init__(self, llm, db_path):
@@ -25,8 +24,8 @@ def format_chat_prompt(self, message, instruction):
         return prompt
 
     def queryblock(self, message):
-        search_result = self.jsonsearch.search_code_contents_by_name(self.db_path, message)
-        return search_result
+        search_result,md = self.jsonsearch.search_code_contents_by_name(self.db_path, message)
+        return search_result,md
 
     def list_to_markdown(self, search_result):
         markdown_str = ""
@@ -38,14 +37,14 @@ def list_to_markdown(self, search_result):
         return markdown_str
 
     def nerquery(self, message):
-        instruction = """
+        query1 = """
         The output must strictly be a pure function name or class name, without any additional characters.
         For example:
         Pure function names: calculateSum, processData
         Pure class names: MyClass, DataProcessor
         The output function name or class name should be only one.
         """
-        query = f"{instruction}\nExtract the most relevant class or function from the following input:\n{message}\nOutput:"
+        query = f"Extract the most relevant class or function base following instrcution {query1},here is input:\n{message}\nOutput:"
         response = self.llm.complete(query)
         # logger.debug(f"Input: {message}, Output: {response}")
         return response
@@ -56,4 +55,4 @@ def nerquery(self, message):
     log_file = "your_logfile_path"
     llm = OpenAI(api_key=api_key, api_base=api_base)
     db_path = "your_database_path"
-    test = TextAnalysisTool(llm, db_path)
+    test = TextAnalysisTool(llm, db_path)

repo_agent/chat_with_repo/rag.py

@@ -4,8 +4,9 @@
 from repo_agent.log import logger
 from llama_index import PromptTemplate
 from llama_index.llms import OpenAI
+import json
+from openai import OpenAI as AI
 
-# logger.add("./log.txt", level="DEBUG", format="{time} - {name} - {level} - {message}")
 
 class RepoAssistant:
     def __init__(self, api_key, api_base, db_path):
@@ -16,6 +17,7 @@ def __init__(self, api_key, api_base, db_path):
         self.md_contents = []
         self.llm = OpenAI(api_key=api_key, api_base=api_base,model="gpt-3.5-turbo-1106")
         self.client = OpenAI(api_key=api_key, api_base=api_base,model="gpt-4-1106-preview")
+        self.lm = AI(api_key = api_key, base_url = api_base)
         self.textanslys = TextAnalysisTool(self.llm,db_path)
         self.json_data = JsonFileProcessor(db_path)
         self.chroma_data = ChromaManager(api_key, api_base)
@@ -37,10 +39,26 @@ def generate_queries(self, query_str: str, num_queries: int = 4):
         queries = response.text.split("\n")
         return queries
 
+    def rerank(self, query ,docs):
+        response = self.lm.chat.completions.create(
+            model='gpt-4-1106-preview',
+            response_format={"type": "json_object"},
+            temperature=0,
+            messages=[
+            {"role": "system", "content": "You are an expert relevance ranker. Given a list of documents and a query, your job is to determine how relevant each document is for answering the query. Your output is JSON, which is a list of documents.  Each document has two fields, content and score.  relevance_score is from 0.0 to 100.0. Higher relevance means higher score."},
+            {"role": "user", "content": f"Query: {query} Docs: {docs}"}
+            ]
+        )
+        scores = json.loads(response.choices[0].message.content)["documents"]
+        logger.debug(f"scores: {scores}")
+        sorted_data = sorted(scores, key=lambda x: x['relevance_score'], reverse=True)
+        top_5_contents = [doc['content'] for doc in sorted_data[:5]]
+        return top_5_contents
+
     def rag(self, query, retrieved_documents):
         # rag 
         information = "\n\n".join(retrieved_documents)
-        messages = f"You are a helpful expert repo research assistant. Your users are asking questions about information contained in a repository. You will be shown the user's question, and the relevant information from the repository. Answer the user's question using only  the information given.\nQuestion: {query}. \nInformation: {information}"
+        messages = f"You are a helpful expert repo research assistant. Your users are asking questions about information contained in repo . You will be shown the user's question, and the relevant information from the repo. Answer the user's question using only this information.\nQuestion: {query}. \nInformation: {information}"
         response = self.llm.complete(messages)
         content = response
         return content
@@ -49,7 +67,7 @@ def list_to_markdown(self,list_items):
 
         # 对于列表中的每个项目，添加一个带数字的列表项
         for index, item in enumerate(list_items, start=1):
-            markdown_content += f"[{index}] {item}\n"
+            markdown_content += f"{index}. {item}\n"
 
         return markdown_content
     def rag_ar(self, query, related_code, embedding_recall, project_name):
@@ -83,7 +101,7 @@ def respond(self, message, instruction):
         # return answer
         prompt = self.textanslys.format_chat_prompt(message, instruction)
         questions = self.textanslys.keyword(prompt)
-        logger.debug(f"Questions: {questions}")
+        # logger.debug(f"Questions: {questions}")
         promptq = self.generate_queries(prompt,3)
         all_results = []
         all_ids = []
@@ -93,7 +111,8 @@ def respond(self, message, instruction):
             all_ids.extend(query_result['ids'][0])
 
         logger.debug(f"all_ids: {all_ids},{all_results}")
-        unique_ids = [id for id in all_ids if all_ids.count(id) == 1]
+        unique_ids = list(dict.fromkeys(all_ids))
+        # unique_ids = [id for id in all_ids if all_ids.count(id) == 1]
         logger.debug(f"uniqueid: {unique_ids}")
         unique_documents = []
         unique_code = []
@@ -102,30 +121,49 @@ def respond(self, message, instruction):
                 if id in unique_ids:
                     unique_documents.append(doc)
                     unique_code.append(code.get("code_content"))
-        unique_code=self.textanslys.list_to_markdown(unique_code)
-        retrieved_documents = unique_documents
+
+        retrieved_documents = self.rerank(message,unique_documents)
         # logger.debug(f"retrieveddocuments: {retrieved_documents}")
         response = self.rag(prompt,retrieved_documents)
         chunkrecall = self.list_to_markdown(retrieved_documents)
         bot_message = str(response)
         keyword = str(self.textanslys.nerquery(bot_message))
         keywords = str(self.textanslys.nerquery(str(prompt)+str(questions)))
-        codez=self.textanslys.queryblock(keyword)
-        codey=self.textanslys.queryblock(keywords)
+        codez,mdz=self.textanslys.queryblock(keyword)
+        codey,mdy=self.textanslys.queryblock(keywords)
         if not isinstance(codez, list):
-            codex = [codez]
+            codez = [codez]
+        if not isinstance(mdz, list):
+            mdz = [mdz]
         # 确保 codey 是列表，如果不是，则将其转换为列表
         if not isinstance(codey, list):
             codey = [codey]
+        if not isinstance(mdy, list):
+            mdy = [mdy]
+
         codex = codez+codey
-        codex = self.textanslys.list_to_markdown(codex)
-        bot_message = self.rag_ar(prompt,unique_code,retrieved_documents,"test")
-        bot_message = str(bot_message) +'\n'+ str(self.textanslys.tree(bot_message))
-        return message, bot_message,chunkrecall,questions,unique_code,codex
-
+        md = mdz + mdy
+        unique_mdx = list(set([item for sublist in md for item in sublist]))
+        uni_codex = []
+        uni_md = []
+        uni_codex = list(dict.fromkeys(codex))
+        uni_md = list(dict.fromkeys(unique_mdx))
+        codex = self.textanslys.list_to_markdown(uni_codex)
+        retrieved_documents = retrieved_documents+uni_md
+        retrieved_documents = list(dict.fromkeys(retrieved_documents))
+        retrieved_documents = self.rerank(message,retrieved_documents[:6])
+        uni_code = uni_codex+unique_code
+        uni_code = list(dict.fromkeys(uni_code))
+        uni_code = self.rerank(message,uni_code[:6])
+        unique_code=self.textanslys.list_to_markdown(unique_code)
+        bot_message = self.rag_ar(prompt,uni_code,retrieved_documents,"test")
+        bot_message = str(bot_message)
+        return message, bot_message, chunkrecall, questions, unique_code, codex
+
+
 if __name__ == "__main__":
     api_key = ""
     api_base = ""
     db_path = ""
     log_file = ""
-    assistant = RepoAssistant(api_key, api_base, db_path, log_file)
+    assistant = RepoAssistant(api_key, api_base, db_path, log_file)