VoidZeroe/scouter

1 2import gradio as gr3import random4import time5import os6import requests7from langchain.document_loaders import TextLoader  #for textfiles8from langchain.text_splitter import CharacterTextSplitter #text splitter9from langchain.embeddings.openai import OpenAIEmbeddings #for using HugginFace models10from langchain.vectorstores import FAISS  #facebook vectorizationfrom langchain.chains.question_answering import load_qa_chain11from langchain.chains.question_answering import load_qa_chain12from langchain import OpenAI13from langchain.document_loaders import UnstructuredPDFLoader  #load pdf14from langchain.indexes import VectorstoreIndexCreator #vectorize db index with chromadb15from langchain.chains import RetrievalQAWithSourcesChain16from langchain.document_loaders import DirectoryLoader17from langchain.document_loaders import UnstructuredFileLoader18 19from langchain.document_loaders.csv_loader import CSVLoader20from langchain.document_loaders import TextLoader21from langchain.document_loaders import Docx2txtLoader22 23from langchain.vectorstores import FAISS24 25from langchain.callbacks.base import BaseCallbackHandler26 27from openai.error import APIError28 29from langchain.embeddings.openai import OpenAIEmbeddings30from langchain.chat_models import ChatOpenAI31from langchain.vectorstores import FAISS32from langchain.chains import LLMChain33from langchain.chains.conversational_retrieval.prompts import CONDENSE_QUESTION_PROMPT34from langchain.prompts import MessagesPlaceholder35from langchain.callbacks.base import BaseCallbackHandler36 37from langchain.prompts import PromptTemplate38from langchain.prompts import ChatPromptTemplate39 40 41openAI_embeddings = OpenAIEmbeddings(openai_api_key = "sk***2T")42 43 44 45 46loader = CSVLoader(file_path='survey.csv', csv_args={47    'delimiter': ',',48    'quotechar': '"',49    'fieldnames': ['Question', 'Answer']50})51 52docs = loader.load()53char_text_splitter = CharacterTextSplitter(chunk_size = 500, chunk_overlap = 15)54docs = char_text_splitter.split_documents(docs)55surveydb = FAISS.from_documents(docs, openAI_embeddings)56 57 58 59loader = TextLoader("transcript.txt")60 61 62docs = loader.load()63char_text_splitter = CharacterTextSplitter(chunk_size = 500, chunk_overlap = 15)64docs = char_text_splitter.split_documents(docs)65transcriptdb = FAISS.from_documents(docs, openAI_embeddings)66 67loader = TextLoader("virscout.txt")68 69docs = loader.load()70char_text_splitter = CharacterTextSplitter(chunk_size = 500, chunk_overlap = 15)71docs = char_text_splitter.split_documents(docs)72virscoutdb = FAISS.from_documents(docs, openAI_embeddings)73 74loader = TextLoader("evaluation.txt")75 76docs = loader.load()77char_text_splitter = CharacterTextSplitter(chunk_size = 500, chunk_overlap = 15)78docs = char_text_splitter.split_documents(docs)79evaluatedb = FAISS.from_documents(docs, openAI_embeddings)80 81loader = Docx2txtLoader("note.docx")82 83docs = loader.load()84char_text_splitter = CharacterTextSplitter(chunk_size = 500, chunk_overlap = 15)85docs = char_text_splitter.split_documents(docs)86notedb = FAISS.from_documents(docs, openAI_embeddings)87 88from queue import Queue, Empty89 90class QueueCallback(BaseCallbackHandler):91    """Callback handler for streaming LLM responses to a queue."""92 93    def __init__(self, q):94        self.q = q95 96    def on_llm_new_token(self, token: str, **kwargs) -> None:97        self.q.put(token)98 99    def on_llm_end(self, *args, **kwargs) -> None:100        return self.q.empty()101 102 103def stream_docs(input_text,104                apikey, 105                chat_history = [], 106                role_grade = 7, 107                survey_vectorstore = surveydb, 108                note_vectorstore = notedb, 109                evaluation_vectorstore = evaluatedb, 110                virscout_vectorstore = virscoutdb, 111                interview_vectorstore = transcriptdb):112 113    # Create a Queue114    q = Queue()115    job_done = object()116 117    system_message = """Use the information from the below sources to answer any questions.118    Role: You are a scout that helps in deciding if a player will be a good fit for a team.119 120    The provided data is about a baseball player to be drafted.121 122    Survey: Survey of medical questions and responses and general knowledge about the player123    <survey>124    {survey}125    </survey>126 127    Scouts' Note: Notes about the player128        The note is in this format:129            The date, followed by the scout name.130            The next line is the scout title131            The third line is the scout note about the player132    <scouts note>133    {scouts_note}134    </scouts note>135 136    Scouts' Evaluation Note: Evaluation Notes about the player137        The note is in this format:138            The first line is the scout last name139            The second line is the scout evaluation of the player140    <evaluation note>141    {evaluation_note}142    </evaluation note>143 144    VirScout: Virscout Interview Notes and Mindset data about the player145        The note is in this format:146            Multiple interview question answer pair147            Then followed by mindset data148            The mindset data contains a key called Pro Sim, this measures the similarity of the player to an ideal pro player, it is a value that ranges from -1 to 1149    <virscout>150    {virscout}151    </virscout>152 153    Scout Interview: Transcript of the player interview with a scout.154        This transcript consists only of words uttered by the player.155    <interview>156    {interview}157    </interview>158 159    Role Grade: The role grade given to the player by different scouts160        The text is in this format:161            Scout Name: Role grade162        163        Use the following definitions to interprete the Role grade in the form `Role grade: Interpretation`164            8: Hall of Fame / Elite165            7: Consistent All-Star166            6: All-Star167            5.5: Solid Everyday168            5: Everyday169            4.5: Platoon / Bench170            4: Up /Down171            3: Org Player172            2: Non-Prospect173    <role_grade>174    {role_grade}175    </role_grade>176 177    Use the following piece of context, if you don't know the answer, simply say I don't know. Always give the answer in Markdown.178 179    {context}180    """181 182    llm = ChatOpenAI(temperature=0, model_name='gpt-4', openai_api_key = apikey)183 184    llm_chain = LLMChain(185        llm=llm,186        prompt=CONDENSE_QUESTION_PROMPT187    )188 189    standalone_question = llm_chain(190        {"chat_history": chat_history, "question": input_text})['text']191    192    qa_prompt = ChatPromptTemplate.from_messages(193        [("system", system_message), ("human", "{question}")])194 195    llm = ChatOpenAI(streaming=True, callbacks=[QueueCallback(196        q)], temperature=0, model_name='gpt-3.5-turbo', 197                     openai_api_key = apikey)198    199    # Initialize the LLM we'll be using200    full_chain = {201        "survey": (lambda x: x['question']) | survey_vectorstore.as_retriever(),202        "interview": (lambda x: x['question']) | interview_vectorstore.as_retriever(),203        "scouts_note": (lambda x: x['question']) | note_vectorstore.as_retriever(),204        "virscout": (lambda x: x['question']) | virscout_vectorstore.as_retriever(),205        "evaluation_note": (lambda x: x['question']) | evaluation_vectorstore.as_retriever(),206        "role_grade": (lambda x: x['role_grade']),207        "question": lambda x: x['question'],208        "context": lambda x: x['chat_history']209    } | qa_prompt | llm210 211    # Create a function to call - this will run in a thread212    return(full_chain.invoke({"question": standalone_question,213                          "chat_history": chat_history,214                          "role_grade":role_grade}))215 216    217 218 219with gr.Blocks(theme= gr.themes.Monochrome()) as demo:220    with gr.Group():221        gr.Label("ScoutBot", show_label=False)222        apikey = gr.Textbox(show_label=False, type="password", placeholder="OpenAI Key")223        chatbot = gr.Chatbot(layout="bubble",bubble_full_width=False, label = "Test Suite")224        msg = gr.Textbox(show_label = False, placeholder = "Type a message")225        with gr.Row():226            submit = gr.Button("Submit")227            clear = gr.ClearButton([msg, chatbot])228 229    def respond(message, chat_history, apikey):230        try:231            bot_message = stream_docs(message, apikey, chat_history=chat_history).content232        except:233            bot_message = "please verify your api key above"234        chat_history.append((message, bot_message))235        #time.sleep(2)236        return "", chat_history237 238    msg.submit(respond, [msg, chatbot, apikey], [msg, chatbot])239    submit.click(respond, [msg, chatbot, apikey], [msg, chatbot])240demo.launch(share = True)241