src/functions_llm.py

from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
import streamlit as st
import sentry_sdk
from sentry_sdk.integrations.serverless import serverless_function
from dotenv import load_dotenv
import os

load_dotenv()

sentry_dsn = os.getenv("SENTRY_DSN")

sentry_sdk.init(
    dsn=sentry_dsn,
    send_default_pii=True,
    traces_sample_rate=1.0,
    _experiments={
        "continuous_profiling_auto_start": True,
    },
)

@serverless_function
@st.cache_resource
def load_model(model_name):
    try:
        # Load model and tokenizer from HFHub
        model = AutoModelForSequenceClassification.from_pretrained(model_name)
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        sentiment_analyzer = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
        return sentiment_analyzer
    except Exception as e:
        st.error(f"Error loading model: {e}")
        return None

@serverless_function
def compare_label_results(df, column_with_predicted_result, column_with_real_result):
    real_results = df[column_with_real_result].tolist()
    predicted_results = df[column_with_predicted_result].tolist()
    correct = 0
    for real, predicted in zip(real_results, predicted_results):
        if real.lower() == predicted.lower():
            correct += 1
    return correct / len(real_results)