import torch
import torch.nn as nn
from transformers import AutoTokenizer, AutoModel
import gradio as gr

# Model multitask (Topik & Sentimen)
class MultiTaskModel(nn.Module):
    def __init__(self, base_model_name, num_topic_classes, num_sentiment_classes):
        super(MultiTaskModel, self).__init__()
        self.encoder = AutoModel.from_pretrained(base_model_name)
        hidden_size = self.encoder.config.hidden_size
        self.topic_classifier = nn.Linear(hidden_size, num_topic_classes)
        self.sentiment_classifier = nn.Linear(hidden_size, num_sentiment_classes)

    def forward(self, input_ids, attention_mask):
        outputs = self.encoder(input_ids=input_ids, attention_mask=attention_mask)
        pooled_output = outputs.last_hidden_state[:, 0]
        topic_logits = self.topic_classifier(pooled_output)
        sentiment_logits = self.sentiment_classifier(pooled_output)
        return topic_logits, sentiment_logits

# Load tokenizer & model
tokenizer = AutoTokenizer.from_pretrained("tokenizer")
model = MultiTaskModel("indobenchmark/indobert-base-p1", num_topic_classes=4, num_sentiment_classes=3)
model.load_state_dict(torch.load("model.pt", map_location=torch.device("cpu")))
model.eval()

# Label mapping
topic_labels = ["Produk", "Layanan", "Pengiriman", "Lainnya"]
sentiment_labels = ["Negatif", "Netral", "Positif"]

# Fungsi klasifikasi
def klasifikasi(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        topic_logits, sentiment_logits = model(**inputs)
        topic_probs = torch.softmax(topic_logits, dim=-1).squeeze()
        sentiment_probs = torch.softmax(sentiment_logits, dim=-1).squeeze()

    topic_result = {label: float(prob) for label, prob in zip(topic_labels, topic_probs)}
    sentiment_result = {label: float(prob) for label, prob in zip(sentiment_labels, sentiment_probs)}
    return {"Topik": topic_result, "Sentimen": sentiment_result}

# Gradio UI
demo = gr.Interface(fn=klasifikasi, inputs="text", outputs="json", title="Klasifikasi Topik dan Sentimen Pelanggan")
demo.launch()