ai_generate.py

import torch
from openai import OpenAI
import os
from transformers import pipeline
from groq import Groq
import base64
import vertexai
from vertexai.generative_models import GenerativeModel, Part, FinishReason
import vertexai.preview.generative_models as generative_models
import google.generativeai as genai
import anthropic
from langchain_community.document_loaders import PyMuPDFLoader
from langchain_community.document_loaders import TextLoader
from langchain_community.embeddings.sentence_transformer import (
    SentenceTransformerEmbeddings,
)
from langchain_community.vectorstores import Chroma
from langchain_text_splitters import CharacterTextSplitter
from langchain import hub
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
from langchain.chains import RetrievalQA
from langchain_groq import ChatGroq
from dotenv import load_dotenv

load_dotenv()

os.environ["GRPC_VERBOSITY"] = "ERROR"
os.environ["GLOG_minloglevel"] = "2"

groq_client = Groq(
    api_key=os.environ.get("GROQ_API_KEY"),
)
openai_client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
# give access to all APIs for GCP instance
# gcloud auth application-default login
genai.configure(api_key=os.environ.get("GENAI_API_KEY"))
vertexai.init(project="proprietary-info-detection", location="us-central1")
gemini_client = GenerativeModel("gemini-1.5-pro-001")
claude_client = anthropic.Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY"))


def create_db_with_langchain(path):
    loader = PyMuPDFLoader(path)
    data = loader.load()
    # split it into chunks
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    docs = text_splitter.split_documents(data)

    # create the open-source embedding function
    embedding_function = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")

    # load it into Chroma
    db = Chroma.from_documents(docs, embedding_function)
    return db


def generate_groq_rag(text, model, path):
    llm = ChatGroq(
        temperature=0,
        model_name=model,
    )
    db = create_db_with_langchain(path)
    retriever = db.as_retriever(search_type="mmr", search_kwargs={"k": 4, "fetch_k": 20})
    prompt = hub.pull("rlm/rag-prompt")

    def format_docs(docs):
        return "\n\n".join(doc.page_content for doc in docs)

    rag_chain = {"context": retriever | format_docs, "question": RunnablePassthrough()} | prompt | llm
    return rag_chain.invoke(text).content


def generate_groq_base(text, model):
    completion = groq_client.chat.completions.create(
        model=model,
        messages=[
            {"role": "user", "content": text},
            {
                "role": "assistant",
                "content": "Please follow the instruction and write about the given topic in approximately the given number of words",
            },
        ],
        temperature=1,
        max_tokens=1024,
        stream=True,
        stop=None,
    )
    response = ""
    for i, chunk in enumerate(completion):
        if i != 0:
            response += chunk.choices[0].delta.content or ""
    return response


def generate_groq(text, model, path):
    if path:
        return generate_groq_rag(text, model, path)
    else:
        return generate_groq_base(text, model)


def generate_openai(text, model, openai_client):
    message = [{"role": "user", "content": text}]
    response = openai_client.chat.completions.create(
        model=model,
        messages=message,
        temperature=1,
        max_tokens=1024,
    )
    return response.choices[0].message.content


def generate_gemini(text, model, gemini_client):
    safety_settings = {
        generative_models.HarmCategory.HARM_CATEGORY_HATE_SPEECH: generative_models.HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
        generative_models.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: generative_models.HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
        generative_models.HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: generative_models.HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
        generative_models.HarmCategory.HARM_CATEGORY_HARASSMENT: generative_models.HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
    }
    generation_config = {
        "max_output_tokens": 1024,
        "temperature": 1.0,
        "top_p": 1.0,
    }
    response = gemini_client.generate_content(
        [text],
        generation_config=generation_config,
        safety_settings=safety_settings,
        stream=False,
    )
    return response.text


def generate_claude(text, model, claude_client):
    response = claude_client.messages.create(
        model=model,
        max_tokens=1024,
        temperature=1.0,
        system="You are helpful assistant.",
        messages=[{"role": "user", "content": [{"type": "text", "text": text}]}],
    )
    return response.content[0].text.strip()


def generate(text, model, path, api=None):

    if model == "LLaMA 3":
        return generate_groq(text, "llama3-70b-8192", path)
    elif model == "OpenAI GPT 4o Mini":
        return generate_openai(text, "gpt-4o-mini", openai_client)
    elif model == "OpenAI GPT 4o":
        return generate_openai(text, "gpt-4o", openai_client)
    elif model == "OpenAI GPT 4":
        return generate_openai(text, "gpt-4-turbo", openai_client)
    elif model == "Gemini 1.5 Pro":
        return generate_gemini(text, "", gemini_client)
    elif model == "Claude Sonnet 3.5":
        return generate_claude(text, "claude-3-5-sonnet-20240620", claude_client)