from transformers import (
    pipeline,
    AutoTokenizer,
    AutoModelForSequenceClassification,
)
import os

CACHE_ROOT = os.environ.get("HF_CACHE_DIR", "/tmp/hf-cache")

os.makedirs(CACHE_ROOT, exist_ok=True)
os.environ["HF_HOME"] = CACHE_ROOT
os.environ["TRANSFORMERS_CACHE"] = os.path.join(CACHE_ROOT, "transformers")
os.environ["HF_HUB_CACHE"] = os.path.join(CACHE_ROOT, "hub")

def predict_sentiment(text: str, task: str, model_name: str):
    cache_dir = os.environ.get("TRANSFORMERS_CACHE") 
    tokenizer = AutoTokenizer.from_pretrained(model_name, cache_dir=cache_dir, use_fast=True)
    model = AutoModelForSequenceClassification.from_pretrained(model_name, cache_dir=cache_dir)

    pipe = pipeline(task = task, 
                    model = model,
                    tokenizer=tokenizer,
                    
                   )

    pipeline_result = pipe(text) # замените на выход пайплайна!
    return pipeline_result