Upload 3 files

app.py

@@ -0,0 +1,74 @@
+import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoConfig
+from huggingface_hub import hf_hub_url
+import os
+
+# Impor kelas kustom Anda secara eksplisit
+from model import IndoBERTClassifier 
+
+# --- Konfigurasi dan Pemuatan Model ---
+MODEL_ID = "TrioF/KlikBERT" 
+
+# Muat tokenizer dan config dari Hub
+config = AutoConfig.from_pretrained(MODEL_ID)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+
+# Inisialisasi kelas kustom dan muat bobot dari Hub
+model = IndoBERTClassifier(config)
+model_path = hf_hub_url(repo_id=MODEL_ID, filename="pytorch_model.bin")
+model.load_state_dict(torch.hub.load_state_dict_from_url(model_path, map_location="cpu"))
+model.eval()
+
+# --- Pemetaan Label ---
+# Pastikan config.json Anda sudah menggunakan 'custom_id2label'
+id2label_clickbait = config.custom_id2label['clickbait']
+id2label_kategori = config.custom_id2label['kategori']
+
+
+# --- Fungsi Prediksi ---
+def predict(judul, isi):
+    inputs = tokenizer(
+        judul,
+        isi,
+        truncation=True,
+        padding=True,
+        max_length=512,
+        return_tensors="pt"
+    )
+    
+    with torch.no_grad():
+        outputs = model(**inputs)
+    
+    clickbait_logits = outputs["clickbait_logits"]
+    kategori_logits = outputs["kategori_logits"]
+    
+    pred_clickbait_id = torch.argmax(clickbait_logits, dim=1).item()
+    pred_kategori_id = torch.argmax(kategori_logits, dim=1).item()
+    
+    pred_clickbait_label = id2label_clickbait[str(pred_clickbait_id)]
+    pred_kategori_label = id2label_kategori[str(pred_kategori_id)]
+    
+    # --- PERUBAHAN DI SINI ---
+    # Kembalikan dua nilai terpisah, bukan dictionary
+    return pred_clickbait_label, pred_kategori_label
+
+
+# --- Antarmuka Gradio ---
+inputs = [
+    gr.Textbox(lines=2, label="Judul Berita", placeholder="Masukkan judul berita di sini..."),
+    gr.Textbox(lines=10, label="Isi Berita", placeholder="Masukkan isi berita di sini...")
+]
+
+# --- PERUBAHAN DI SINI ---
+# Gunakan dua komponen output terpisah
+outputs = [
+    gr.Text(label="Prediksi Clickbait"),
+    gr.Text(label="Prediksi Kategori Berita")
+]
+
+title = "Model Multi-Task KlikBERT"
+description = "Model ini memprediksi apakah judul clickbait dan apa kategori beritanya. Model ini dimuat dari repositori TrioF/KlikBERT."
+    
+iface = gr.Interface(fn=predict, inputs=inputs, outputs=outputs, title=title, description=description)
+iface.launch()

model.py

@@ -0,0 +1,33 @@
+# Nama file: model.py
+
+import torch
+import torch.nn as nn
+from transformers import AutoModel, AutoConfig
+
+class IndoBERTClassifier(nn.Module):
+    def __init__(self, config):
+        super(IndoBERTClassifier, self).__init__()
+        # Gunakan config dari model dasar untuk mengambil hidden_size
+        self.bert = AutoModel.from_pretrained(config._name_or_path, config=config)
+        self.dropout = nn.Dropout(config.classifier_dropout if hasattr(config, 'classifier_dropout') else 0.1)
+        hidden_size = self.bert.config.hidden_size
+
+        self.num_clickbait_labels = config.num_clickbait_labels
+        self.num_kategori_labels = config.num_kategori_labels
+
+        self.clickbait_classifier = nn.Linear(hidden_size, self.num_clickbait_labels)
+        self.kategori_classifier = nn.Linear(hidden_size, self.num_kategori_labels)
+
+    def forward(self, input_ids, attention_mask, clickbait_labels=None, kategori_labels=None, **kwargs):
+        output = self.bert(input_ids=input_ids, attention_mask=attention_mask)
+        pooled_output = output.last_hidden_state[:, 0, :]  # Ambil token [CLS]
+
+        dropout_output = self.dropout(pooled_output)
+
+        clickbait_logits = self.clickbait_classifier(dropout_output)
+        kategori_logits = self.kategori_classifier(dropout_output)
+
+        return {
+            "clickbait_logits": clickbait_logits,
+            "kategori_logits": kategori_logits
+        }

requirements.txt

@@ -0,0 +1,4 @@
+torch
+transformers
+scikit-learn
+gradio