import torch
from transformers import BertTokenizer, BertModel, T5ForConditionalGeneration
from transformers import AutoModelForSequenceClassification
from utilities import preprocess_features, generate_recommendation

MODEL_PATH = "your-hf-username/deal-score-model"  # Replace with your model path
SUMMARIZER_PATH = "t5-small"  # Can use distilgpt2 or other summarizers

def load_model():
    model = AutoModelForSequenceClassification.from_pretrained(MODEL_PATH)
    tokenizer = BertTokenizer.from_pretrained(MODEL_PATH)
    summarizer = T5ForConditionalGeneration.from_pretrained(SUMMARIZER_PATH)
    return model, tokenizer, summarizer

def predict_deal_score(input_data, model, tokenizer, summarizer):
    features = preprocess_features(input_data)
    inputs = tokenizer(features["text"], return_tensors="pt")
    
    with torch.no_grad():
        outputs = model(**inputs)
        score = round(outputs.logits[0][0].item() * 100)
        confidence = torch.softmax(outputs.logits[0], dim=-1).max().item()

    risk = (
        "Low" if score > 75 else
        "Medium" if score > 50 else
        "High"
    )

    recommendation = generate_recommendation(summarizer, input_data)

    return {
        "score": score,
        "confidence": round(confidence, 2),
        "risk": risk,
        "recommendation": recommendation
    }