app.py

import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch.nn.functional as F

# =========================
# Model
# =========================
MODEL_ID = "oddadmix/dialect-router-v0.1"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
model.eval()

# =========================
# Label Mapping
# =========================
LABEL_MAP = {
    "eg": {"dialect": "Egyptian", "region": "Egypt", "flag": "🇪🇬"},
    "sa": {"dialect": "Saudi", "region": "Saudi Arabia", "flag": "🇸🇦"},
    "mo": {"dialect": "Moroccan (Darija)", "region": "Morocco", "flag": "🇲🇦"},
    "iq": {"dialect": "Iraqi", "region": "Iraq", "flag": "🇮🇶"},
    "sd": {"dialect": "Sudanese", "region": "Sudan", "flag": "🇸🇩"},
    "tn": {"dialect": "Tunisian", "region": "Tunisia", "flag": "🇹🇳"},
    "lb": {"dialect": "Lebanese", "region": "Lebanon", "flag": "🇱🇧"},
    "sy": {"dialect": "Syrian", "region": "Syria", "flag": "🇸🇾"},
    "ly": {"dialect": "Libyan", "region": "Libya", "flag": "🇱🇾"},
    "ps": {"dialect": "Palestinian", "region": "Palestine", "flag": "🇵🇸"},
    "ar": {"dialect": "Modern Standard Arabic (MSA)", "region": "—", "flag": "🌍"},
}

# Reverse mapping if model uses numeric labels
id2label = model.config.id2label


def format_label(code):
    meta = LABEL_MAP.get(code, {})
    return f"{meta.get('flag','')} {meta.get('dialect','Unknown')} ({meta.get('region','')})"


def predict_dialect(text):
    if not text.strip():
        return "Please enter text", {}

    inputs = tokenizer(
        text,
        return_tensors="pt",
        truncation=True,
        padding=True
    ).to(device)

    with torch.no_grad():
        outputs = model(**inputs)
        probs = F.softmax(outputs.logits, dim=-1)[0]

    # Build scores
    scores = {}
    for i, prob in enumerate(probs):
        code = id2label[i]  # e.g., "eg", "sa"
        scores[code] = float(prob)

    # Sort
    sorted_scores = dict(sorted(scores.items(), key=lambda x: x[1], reverse=True))

    # Top prediction
    top_code = list(sorted_scores.keys())[0]
    top_label = format_label(top_code)

    # Format scores nicely
    formatted_scores = {
        format_label(code): round(score, 4)
        for code, score in sorted_scores.items()
    }

    return top_label, formatted_scores


# =========================
# UI
# =========================
with gr.Blocks() as demo:
    gr.Markdown("# 🌍 Arabic Dialect Router")
    gr.Markdown("Detect the dialect of Arabic text with region-aware labels.")

    text_input = gr.Textbox(
        label="Enter Arabic text",
        placeholder="مثال: انا رايح الشغل دلوقتي"
    )

    predict_btn = gr.Button("Predict")

    label_output = gr.Textbox(label="Predicted Dialect")
    scores_output = gr.JSON(label="All Dialect Scores")

    gr.Examples(
        examples=[
            "انا رايح الشغل دلوقتي",   # Egyptian
            "شنو درتي اليوم",        # Moroccan
            "كيفك اليوم",            # Levant
            "وش تسوي الحين",         # Saudi
            "ماذا تفعل الآن",        # MSA
        ],
        inputs=text_input
    )

    predict_btn.click(
        fn=predict_dialect,
        inputs=text_input,
        outputs=[label_output, scores_output]
    )

demo.launch()