initial commit

agent.py

@@ -0,0 +1,104 @@
+from dotenv import load_dotenv
+from dotenv import dotenv_values
+import os
+from tqdm.auto import tqdm
+import pinecone
+from langchain.embeddings import OpenAIEmbeddings
+from pinecone_text.sparse import BM25Encoder
+from langchain.retrievers import PineconeHybridSearchRetriever
+from langchain.chat_models import ChatOpenAI
+from langchain.agents import initialize_agent, Tool
+from langchain.tools.base import BaseTool
+from langchain.agents import AgentType
+from langchain.agents.react.base import DocstoreExplorer
+from langchain import LLMMathChain
+from typing import Union
+from langchain.memory import ConversationBufferWindowMemory
+
+
+load_dotenv()
+config = dotenv_values(".env")
+
+
+
+class CalculatorTool(BaseTool):
+    name = "CalculatorTool"
+    
+    description = """
+    Useful for when you need to execute specific math calculations.
+    This tool is only for math calculations and nothing else.
+    Formulate the input as python code.
+    """
+
+    def _run(self, question: str):
+        return exec(question)
+    
+    def _arun(self, value: Union[int, float]):
+        raise NotImplementedError("This tool does not support async")
+
+
+
+class QMLAgent():
+
+    def __init__(self):
+
+        pinecone.init(api_key=config["PINECONE_API_KEY"], environment=config["PINECONE_REGION"])
+
+        index = pinecone.Index(config["INDEX_NAME"])
+
+        embeddings = OpenAIEmbeddings()
+
+
+        bm25_encoder = BM25Encoder()
+        bm25_encoder.load(config["BM25_FILENAME"])
+
+        retriever = PineconeHybridSearchRetriever(
+            embeddings=embeddings, 
+            sparse_encoder=bm25_encoder, 
+            index=index, 
+            top_k=config["TOP_K"])
+
+        llm = ChatOpenAI(model_name=config["CHAT_MODEL"])
+
+        math_tool = CalculatorTool()
+
+        tools = [
+            Tool(
+                name="Search",
+                func=retriever.get_relevant_documents,
+                description="You have to use this to search for knowledge about quantum computing and quantum machine learning.",
+            ),
+            Tool.from_function(
+                name="Match calculation",
+                func=math_tool._run,
+                description="""
+            Useful for when you need to execute specific math calculations.
+            This tool is only for math calculations and nothing else.
+            Formulate the input as python code, always use explicit printing for results!.
+            """
+                #return_direct=False
+
+            ),
+        ]
+
+        memory = ConversationBufferWindowMemory(k=config["MEMORY_LENGTH"], memory_key="chat_history", return_messages=True)
+
+
+        self.agent_chain = initialize_agent(
+            tools, 
+            llm, 
+            agent=AgentType.CHAT_CONVERSATIONAL_REACT_DESCRIPTION, 
+            verbose=True,
+            return_intermediate_steps=False, 
+            memory=memory, 
+            handle_parsing_errors="Always provide only code answer in a single block parseable in JSON, nothing more! Delete your other remarks, just output the pure code! Always produce runnable code, no parts left blank!"
+        )
+
+    def run(self, question):
+        return self.agent_chain.run(question)
+
+if __name__ == '__main__':
+    agent = QMLAgent()
+    question = "What is eigenvector for the matrix [[1,2,3],[4,5,6],[7,8,9]] raised to the second power?"
+
+    print(agent.run(question))

app.py

@@ -0,0 +1,40 @@
+import gradio as gr
+from dotenv import dotenv_values
+
+from agent import QMLAgent
+
+
+config = dotenv_values(".env")
+
+agent = QMLAgent()
+
+
+with gr.Blocks(theme='snehilsanyal/scikit-learn') as demo:
+    #with gr.Tab("QA"):
+    chatbot = gr.Chatbot(label="QML Class Conversation Agent Demo")
+    msg = gr.Textbox()
+    clear = gr.Button("Clear")
+
+    def respond(user_message, chat_history):#, progress=gr.Progress()):
+        global agent
+        bot_message = agent.run(user_message)
+        chat_history.append((user_message, bot_message))
+
+        return "", chat_history
+
+    msg.submit(respond, [msg, chatbot], [msg, chatbot])
+    clear.click(lambda: None, None, chatbot, queue=False)
+
+
+
+db_login = {
+    config["USER_NAME"]: config["USER_PWD"]
+}
+
+def _myauth(username, password):
+    if db_login.get(username) == password:
+        return True
+    return False
+
+#demo.queue(concurrency_count=10).launch(server_port=7860, server_name='0.0.0.0')
+demo.queue(concurrency_count=10).launch(auth=_myauth, server_port=7860, server_name='0.0.0.0')