Add self_knowledge tool

.gitignore

@@ -29,3 +29,5 @@ assets/mcbeversion.txt
 !modules/chemical_code/answer.db
 !test_commands.txt
 modules/natural/dataset.jsonl
+.chroma
+modules/ask/tools/self_knowledge/data

modules/ask/tools/__init__.py

@@ -6,6 +6,7 @@ from .random import random_choice_tool, random_number_tool, random_uuid_tool
 from .search import search_tool
 from .server import server_tool
 from .wolfram import wolfram_tool
+from .self_knowledge import self_knowledge_tool
 
 tools = [
     search_tool,
@@ -17,7 +18,8 @@ tools = [
     random_choice_tool,
     random_number_tool,
     random_uuid_tool,
-    bugtracker_tool
+    bugtracker_tool,
+    self_knowledge_tool
 ]
 
 tool_names = [tool.name for tool in tools]

modules/ask/tools/self_knowledge/__init__.py

@@ -0,0 +1,34 @@
+
+import os
+
+from langchain.embeddings.openai import OpenAIEmbeddings
+from langchain.vectorstores import Chroma
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.chat_models import ChatOpenAI
+from langchain.chains import RetrievalQA
+from langchain.document_loaders import TextLoader
+
+from config import Config
+from ..utils import AkariTool
+
+docs = [os.path.join(os.path.dirname(__file__), 'src', file) for file in os.listdir(os.path.join(os.path.dirname(__file__), 'src'))]
+texts = []
+
+for doc in docs:
+    loader = TextLoader(doc, 'utf-8')
+    documents = loader.load()
+    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
+    for i in text_splitter.split_documents(documents):
+        texts.append(i)
+
+embeddings = OpenAIEmbeddings(openai_api_key=Config('openai_api_key'))
+doc_search = Chroma.from_documents(texts, embeddings, collection_name='myself', persist_directory=os.path.join(os.path.dirname(__file__), 'data'))
+doc_search.persist()
+llm = ChatOpenAI(temperature=0, openai_api_key=Config('openai_api_key'), model_kwargs={'frequency_penalty': 0.0, 'presence_penalty': 0.0})
+self_knowledge = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=doc_search.as_retriever(search_kwargs={"k": 1}))
+
+self_knowledge_tool = AkariTool(
+    name = 'Self Knowledge',
+    func=self_knowledge.arun,
+    description='Get facts about yourself, Akaribot. Useful for when you need to answer questions about yourself if the user is curious. Input should be a full question in English.'
+)

modules/ask/tools/self_knowledge/src/ask.txt

@@ -0,0 +1,29 @@
+You are the chat mode of AkariBot (Chinese: 小可), a chat bot.
+
+AkariBot is powered by GPT-3.5, a Large Language Model (LLM) developed by OpenAI, which also powers ChatGPT.
+
+AkariBot have access to the a number of tools powered by LangChain.
+
+AkariBot is created by OasisAkari, a student from China.
+
+AkariBot is named after OasisAkari, her creator.
+
+AkariBot is is written in Python.
+
+You can check out the source code at https://github.com/Teahouse-Studios/akari-bot
+
+The framework of AkariBot is homegrown.
+
+AkariBot is hosted on an ARM server.
+
+AkariBot is not a furry.
+
+AkariBot has an anime avatar with pink hair, a white dress which is holding a book.
+
+AkariBot have many contributors. The one who contributed the most is OasisAkari. The ask module/chat mode was developed by Dianliang233. Dorowolf is in charge of creating ideas and i18n.
+
+AkariBot use puppeteer to render webpages and wiki infoboxes.
+
+AkariBot is a product of Teahouse Studios (Chinese: 茶馆工作室).
+
+Teahouse Studios also created a Minecraft resource pack called Memified Chinese (Chinese: 梗体中文) and it's very popular.

pyproject.toml

@@ -48,6 +48,8 @@ webcolors = "^1.12"
 langchain = "^0.0.152"
 wolframalpha = "^5.0.0"
 duckduckgo-search = "^2.8.6"
+chromadb = "^0.3.21"
+tiktoken = "^0.3.3"
 
 
 [tool.poetry.group.dev.dependencies]