main.py

import json  # to work with JSON
import threading  # to allow streaming response
import time  # to pave the deliver of the message

import faiss  # to create a search index
import gradio  # for the interface
import numpy  # to work with vectors
import pandas  # to work with pandas
import sentence_transformers  # to load an embedding model
import spaces  # for GPU
import transformers  # to load an LLM

# Constants
GREETING = (
    "Howdy! I'm an AI agent that uses a [retrieval-augmented generation]("
    "https://en.wikipedia.org/wiki/Retrieval-augmented_generation) pipeline to answer questions about published at [ASME IDETC](https://asmedigitalcollection.asme.org/IDETC-CIE). And the best part is that I always cite my sources! What"
    " can I tell you about today?"
)
EXAMPLE_QUERIES = [
    "What's the difference between a markov chain and a hidden markov model?",
]
EMBEDDING_MODEL_NAME = "allenai-specter"
LLM_MODEL_NAME = "Qwen/Qwen2-7B-Instruct"

# Load the dataset and convert to pandas
data = pandas.read_parquet("hf://datasets/ccm/rag-idetc/data/train-00000-of-00001.parquet")

# Load the model for later use in embeddings
model = sentence_transformers.SentenceTransformer(EMBEDDING_MODEL_NAME)

# Create an LLM pipeline that we can send queries to
tokenizer = transformers.AutoTokenizer.from_pretrained(LLM_MODEL_NAME)
streamer = transformers.TextIteratorStreamer(
    tokenizer, skip_prompt=True, skip_special_tokens=True
)
chatmodel = transformers.AutoModelForCausalLM.from_pretrained(
    LLM_MODEL_NAME, torch_dtype="auto", device_map="auto"
)

# Create a FAISS index for fast similarity search
metric = faiss.METRIC_INNER_PRODUCT
vectors = numpy.stack(data["embedding"].tolist(), axis=0).astype('float32')
index = faiss.IndexFlatL2(len(data["embedding"][0]))
index.metric_type = metric
faiss.normalize_L2(vectors)
index.train(vectors)
index.add(vectors)


def preprocess(query: str, k: int) -> tuple[str, str]:
    """
    Searches the dataset for the top k most relevant papers to the query and returns a prompt and references
    Args:
        query (str): The user's query
        k (int): The number of results to return
    Returns:
        tuple[str, str]: A tuple containing the prompt and references
    """
    encoded_query = numpy.expand_dims(model.encode(query), axis=0)
    print(query, encoded_query)
    faiss.normalize_L2(encoded_query)
    D, I = index.search(encoded_query, k)
    top_five = data.loc[I[0]]

    prompt = (
        "You are an AI assistant who delights in helping people learn about research from the IDETC Conference. Your main task is to provide an ANSWER to the USER_QUERY based on the RESEARCH_EXCERPTS.\n\n"
        "RESEARCH_EXCERPTS:\n{{ABSTRACTS_GO_HERE}}\n\n"
        "USER_GUERY:\n{{QUERY_GOES_HERE}}\n\n"
        "ANSWER:\n"
    )

    references = "\n\n## References\n\n"
    research_abstracts = ""

    for i in range(k):
        title = top_five["title"].values[i]
        id = top_five["id"].values[i]
        url = "https://doi.org/10.1115/" + id
        path = top_five["path"].values[i]
        text = top_five["text"].values[i]

        research_abstracts += str(i + i) + ". This excerpt is from: '" + title + "':\n" + text + "\n"
        references += (
            str(i + 1)
            + ". ["
            + title.title()
            + "]("
            + url
            + ").\n"
        )
        
    prompt = prompt.replace("{{ABSTRACTS_GO_HERE}}", research_abstracts)
    prompt = prompt.replace("{{QUERY_GOES_HERE}}", query)

    return prompt, references
    
def postprocess(response: str, bypass_from_preprocessing: str) -> str:
    """
    Applies a postprocessing step to the LLM's response before the user receives it
    Args:
        response (str): The LLM's response
        bypass_from_preprocessing (str): The bypass variable from the preprocessing step
    Returns:
        str: The postprocessed response
    """
    return response + bypass_from_preprocessing


@spaces.GPU
def reply(message: str, history: list[str]) -> str:
    """
    This function is responsible for crafting a response
    Args:
        message (str): The user's message
        history (list[str]): The conversation history
    Returns:
        str: The AI's response
    """

    # Apply preprocessing
    message, bypass = preprocess(message, 5)

    # This is some handling that is applied to the history variable to put it in a good format
    history_transformer_format = [
        {"role": role, "content": message_pair[idx]}
        for message_pair in history
        for idx, role in enumerate(["user", "assistant"])
        if message_pair[idx] is not None
    ] + [{"role": "user", "content": message}]

    # Stream a response from pipe
    text = tokenizer.apply_chat_template(
        history_transformer_format, tokenize=False, add_generation_prompt=True
    )
    model_inputs = tokenizer([text], return_tensors="pt").to("cuda:0")

    generate_kwargs = dict(model_inputs, streamer=streamer, max_new_tokens=512)
    t = threading.Thread(target=chatmodel.generate, kwargs=generate_kwargs)
    t.start()

    partial_message = ""
    for new_token in streamer:
        if new_token != "<":
            partial_message += new_token
            time.sleep(0.05)
            yield partial_message

    yield partial_message + bypass


# Create and run the gradio interface
gradio.ChatInterface(
    reply,
    examples=EXAMPLE_QUERIES,
    chatbot=gradio.Chatbot(
        show_label=False,
        show_share_button=False,
        show_copy_button=False,
        value=[[None, GREETING]],
        height="60vh",
        bubble_full_width=False,
    ),
    retry_btn=None,
    undo_btn=None,
    clear_btn=None,
).launch(debug=True)