Afwa/Binary-SoftwareRequirement-Classification

1import gradio as gr2from transformers import DistilBertForSequenceClassification, DistilBertTokenizerFast3import torch4import re5 6# Load model dan tokenizer dari Hugging Face Hub7model = DistilBertForSequenceClassification.from_pretrained(8    "Afwa/Binary-SoftwareRequirement-DistilBERT-TPE-Model"9)10tokenizer = DistilBertTokenizerFast.from_pretrained(11    "Afwa/Binary-SoftwareRequirement-DistilBERT-TPE-Model"12)13 14# Karena config.json tidak punya id2label, kita buat manual15id2label = {0: "Functional", 1: "Non-Functional"}16 17def is_gibberish(text: str) -> bool:18    """Deteksi input ngaco/random."""19    # Hanya angka/simbol tanpa huruf20    if not re.search(r"[a-zA-Z]", text):21        return True22    # Terlalu pendek (< 3 kata berarti banget)23    if len(text.split()) < 3:24        return True25    return False26 27def predict(text):28    # Cek input kosong29    if not text.strip():30        return "Please enter text"31 32    # Cek input ngaco33    if is_gibberish(text):34        return "Unrecognized input"35 36    # Tokenisasi input37    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)38    39    # Forward pass40    with torch.no_grad():41        outputs = model(**inputs)42 43    # Ambil probabilitas44    probs = torch.nn.functional.softmax(outputs.logits, dim=-1)[0]45 46    # Ambil label top-147    pred_id = torch.argmax(probs).item()48    pred_label = id2label[pred_id]49 50    return pred_label51 52# Gradio interface53demo = gr.Interface(54    fn=predict,55    inputs=gr.Textbox(lines=3, placeholder="Enter your requirements here..."),56    # outputs=gr.Textbox(label="Prediction"),  # cuma tampilkan label57    outputs=gr.Label(),58    title="Binary Classification Software Requirements",59    description="This DistilBERT model (optimized with TPE) classifies requirements into **Functional** or **Non-Functional**. Note: the model is trained on English text only.",60    flagging_mode="never"61)62 63demo.launch()