app.py

import html
import re
from typing import Dict, List, Tuple

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

from grc_utils import lower_grc, normalize_word, heavy, vowel, only_bases

from syllabify import syllabify_joined
from preprocess import process_word

MODEL_OPTIONS: Dict[str, str] = {
    "SyllaMoBert (current)": "Ericu950/SyllaMoBert-grc-macronizer-v1",
    "Macronizer Mini": "Ericu950/macronizer_mini",
}
DEFAULT_MODEL_LABEL = "SyllaMoBert (current)"
DEFAULT_MODEL_ID = MODEL_OPTIONS[DEFAULT_MODEL_LABEL]
MAX_LENGTH = 512

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

_MODEL_CACHE: Dict[str, Tuple[AutoTokenizer, AutoModelForTokenClassification, Dict[int, str]]] = {}


def _get_model_bundle(model_id: str) -> Tuple[AutoTokenizer, AutoModelForTokenClassification, Dict[int, str]]:
    if model_id in _MODEL_CACHE:
        return _MODEL_CACHE[model_id]

    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForTokenClassification.from_pretrained(model_id)
    model.to(device)
    model.eval()
    id2label = model.config.id2label

    _MODEL_CACHE[model_id] = (tokenizer, model, id2label)
    return _MODEL_CACHE[model_id]


def preprocess_greek_line(line: str) -> List[str]:
    # Normalize accents and keep only Greek-letter word spans.
    normalized = normalize_word(line)
    lower = lower_grc(normalized)
    words = lower.split()
    token_lists = [process_word(word) for word in words]
    return [token for tokens in token_lists for token in tokens]


def _normalize_label(raw_label: str) -> int:
    text = raw_label.lower()
    if "long" in text:
        return 1
    if "short" in text:
        return 2
    if text.endswith("_1") or text == "1":
        return 1
    if text.endswith("_2") or text == "2":
        return 2
    return 0


def preprocess_and_syllabify(line: str):
    tokens = preprocess_greek_line(line)
    return syllabify_joined(tokens)


def classify_line(line: str, model_id: str):
    syllables = preprocess_and_syllabify(line)
    if not syllables:
        return []

    tokenizer, model, id2label = _get_model_bundle(model_id)

    encoded = tokenizer(
        syllables,
        is_split_into_words=True,
        return_tensors="pt",
        truncation=True,
        max_length=MAX_LENGTH,
    )

    word_ids = encoded.word_ids(batch_index=0)

    if "token_type_ids" in encoded:
        del encoded["token_type_ids"]

    model_inputs = {k: v.to(device) for k, v in encoded.items()}

    with torch.no_grad():
        outputs = model(**model_inputs)
        probs = F.softmax(outputs.logits, dim=-1)
        predictions = torch.argmax(probs, dim=-1).squeeze(0).cpu().tolist()

    aligned = []
    seen_word_ids = set()

    for i, word_id in enumerate(word_ids):
        if word_id is None:
            continue
        if word_id in seen_word_ids:
            continue
        if word_id >= len(syllables):
            break

        seen_word_ids.add(word_id)
        pred_id = int(predictions[i])
        label_name = id2label.get(pred_id, str(pred_id))
        normalized = _normalize_label(str(label_name))
        aligned.append((syllables[word_id], normalized))

    return aligned


def _syllable_chip(syllable: str, label_id: int) -> str:
        escaped = html.escape(syllable)
        if label_id == 1:
                return f'<span class="chip long">{escaped}<small>long</small></span>'
        if label_id == 2:
                return f'<span class="chip short">{escaped}<small>short</small></span>'
        return f'<span class="chip clear">{escaped}</span>'


def _mark_syllable_plain(syllable: str, label_id: int) -> str:
    if label_id not in (1, 2):
        return syllable

    marker = "_" if label_id == 1 else "^"
    chars = list(syllable)

    for i in range(len(chars) - 1, -1, -1):
        if vowel(chars[i]):
            return "".join(chars[: i + 1]) + marker + "".join(chars[i + 1 :])

    return syllable + marker


def _to_final_sigma(text: str) -> str:
    # Step 3: in rendered output, only word-final sigmas become final-sigma.
    def _convert_word(token: str) -> str:
        if not token.strip():
            return token

        chars = list(token)
        last_greek_idx = -1
        for i, ch in enumerate(chars):
            if "\u0370" <= ch <= "\u03ff" or "\u1f00" <= ch <= "\u1fff":
                last_greek_idx = i

        if last_greek_idx != -1 and chars[last_greek_idx] == "σ":
            chars[last_greek_idx] = "ς"

        return "".join(chars)

    return "".join(_convert_word(tok) for tok in re.findall(r"\S+|\s+", text))


def _restore_expanded_word(marked_word: str, reference_word: str) -> str:
    restored = marked_word.replace("δσ", "ζ").replace("κσ", "ξ").replace("πσ", "ψ")

    ref_norm = lower_grc(normalize_word(reference_word))
    if "ῥ" in ref_norm:
        rho_idx = restored.find("ρ")
        if rho_idx != -1:
            restored = restored[:rho_idx] + "ῥ" + restored[rho_idx + 1 :]

    return _to_final_sigma(restored)


def _consume_word_alignment(
    aligned: List[Tuple[str, int]],
    start_idx: int,
    expected_syllables: List[str],
) -> Tuple[List[Tuple[str, int]], int]:
    if start_idx >= len(aligned):
        return [], start_idx

    expected_bases = only_bases("".join(expected_syllables))
    if expected_bases:
        taken: List[Tuple[str, int]] = []
        i = start_idx
        while i < len(aligned):
            taken.append(aligned[i])
            current_bases = only_bases("".join(s for s, _ in taken))
            if current_bases == expected_bases:
                return taken, i + 1
            if len(current_bases) > len(expected_bases) and not current_bases.startswith(expected_bases):
                break
            i += 1

    fallback_count = len(expected_syllables)
    if fallback_count <= 0:
        return [], start_idx

    end_idx = min(len(aligned), start_idx + fallback_count)
    return aligned[start_idx:end_idx], end_idx


def _render_plain_line_with_spacing(line: str, aligned: List[Tuple[str, int]]) -> str:
    # Step 1: normalize input final sigma to medial sigma for matching only.
    line_for_matching = line.replace("ς", "σ")
    parts = re.findall(r"\S+|\s+", line)
    parts_for_matching = re.findall(r"\S+|\s+", line_for_matching)
    out_parts: List[str] = []
    cursor = 0

    for part, part_for_matching in zip(parts, parts_for_matching):
        if part_for_matching.isspace():
            # Step 2: preserve original spacing exactly.
            out_parts.append(part_for_matching)
            continue

        normalized_word = lower_grc(normalize_word(part_for_matching)).replace("ς", "σ")
        expected_tokens = process_word(normalized_word)
        expected_syllables = syllabify_joined(expected_tokens)

        taken, cursor = _consume_word_alignment(aligned, cursor, expected_syllables)
        if not taken:
            out_parts.append(part_for_matching)
            continue

        marked = "".join(_mark_syllable_plain(syl, label) for syl, label in taken)
        restored = _restore_expanded_word(marked, part)
        out_parts.append(restored)

    if cursor < len(aligned):
        tail = "".join(_mark_syllable_plain(syl, label) for syl, label in aligned[cursor:])
        out_parts.append(_to_final_sigma(tail))

    return "".join(out_parts)


def render_results(text: str, model_label: str):
    lines = [line.strip() for line in text.splitlines() if line.strip()]
    if not lines:
        return "<div class='empty'>Enter one or more Greek lines to classify syllables.</div>", ""

    model_id = MODEL_OPTIONS.get(model_label, DEFAULT_MODEL_ID)

    cards = []
    export_lines = []

    for idx, line in enumerate(lines, start=1):
        aligned = classify_line(line, model_id)
        chips = "".join(_syllable_chip(syl, label) for syl, label in aligned)
        plain_line = _render_plain_line_with_spacing(line, aligned)

        cards.append(
            f"""
            <section class="card">
                    <div class="line-number">Line {idx}</div>
                    <div class="source">{html.escape(line)}</div>
                    <div class="chips">{chips or '<span class="chip clear">(no syllables found)</span>'}</div>
            </section>
            """
        )

        export_lines.append(f"Line {idx}: {line}")
        export_lines.append(f"  {plain_line}" if plain_line else "  (no syllables found)")

    html_result = (
        "<div class='legend'><span class='dot long'></span>Long"
        "<span class='dot short'></span>Short"
        "<span class='dot clear'></span>Unmarked</div>"
        + "".join(cards)
    )

    export_header = [f"Model: {model_label} ({model_id})", ""]
    return html_result, "\n".join(export_header + export_lines)


examples = [
        "νεανίας ἀάατός ἐστιν καὶ καλός. τὰ παῖδες τὰ καλά\nκαλὰ μὲν ἠέξευ, καλὰ δ᾽ ἔτραφες, οὐράνιε Ζεῦ,",
        "Ἆρες, Ἄρες βροτολοιγὲ μιαιφόνε τειχεσιπλῆτα\nἈτρεΐδαι τε καὶ ἄλλοι ἐϋκνήμιδες Ἀχαιοί",
        "ἢ τυφλὸς ἤ τις σκνιπὸς ἢ λέγα βλέπων\nψάμμου θαλασσῶν ἢ σκνιπῶν Αἰγυπτίων",
]


CSS = """
@import url('https://fonts.googleapis.com/css2?family=Cormorant+Garamond:wght@500;600;700&family=Space+Grotesk:wght@400;500;700&display=swap');

:root {
    --bg-start: #0b0b0d;
    --bg-end: #15151b;
    --ink: #f0f0f5;
    --long: #ff7868;
    --short: #66dbd8;
    --clear: #a0a0ab;
    --paper: rgba(22, 22, 28, 0.9);
    --chip-long-color: var(--long);
    --chip-short-color: var(--short);
    --chip-clear-color: #c9c9d3;
    --source-text: var(--ink);
}

@media (prefers-color-scheme: dark) {
    :root {
        --bg-start: #050506;
        --bg-end: #101015;
        --ink: #f3f3f8;
        --long: #ff7f70;
        --short: #69e2de;
        --clear: #b5b5c2;
        --paper: rgba(16, 16, 22, 0.94);
        --chip-long-color: #ff9b8d;
        --chip-short-color: #7cebe7;
        --chip-clear-color: #d4d4de;
        --source-text: #fcfcff;
    }
    body.dark-mode {
        --bg-start: #050506;
        --bg-end: #101015;
        --ink: #f3f3f8;
        --long: #ff7f70;
        --short: #69e2de;
        --clear: #b5b5c2;
        --paper: rgba(16, 16, 22, 0.94);
        --chip-long-color: #ff9b8d;
        --chip-short-color: #7cebe7;
        --chip-clear-color: #d4d4de;
        --source-text: #fcfcff;
    }
}

body.dark-mode {
    --bg-start: #050506;
    --bg-end: #101015;
    --ink: #f3f3f8;
    --long: #ff7f70;
    --short: #69e2de;
    --clear: #b5b5c2;
    --paper: rgba(16, 16, 22, 0.94);
    --chip-long-color: #ff9b8d;
    --chip-short-color: #7cebe7;
    --chip-clear-color: #d4d4de;
    --source-text: #fcfcff;
}

html.dark-mode {
    --bg-start: #050506;
    --bg-end: #101015;
    --ink: #f3f3f8;
    --long: #ff7f70;
    --short: #69e2de;
    --clear: #b5b5c2;
    --paper: rgba(16, 16, 22, 0.94);
    --chip-long-color: #ff9b8d;
    --chip-short-color: #7cebe7;
    --chip-clear-color: #d4d4de;
    --source-text: #fcfcff;
}

.gradio-container {
    font-family: 'Space Grotesk', sans-serif;
    background: radial-gradient(circle at top left, var(--bg-start), var(--bg-end));
    color: var(--ink);
    transition: background-color 0.3s, color 0.3s;
}

.dark-mode-toggle {
    position: fixed;
    top: 20px;
    right: 20px;
    background: var(--paper);
    border: 2px solid var(--ink);
    color: var(--ink);
    padding: 0.6rem 1.2rem;
    border-radius: 999px;
    cursor: pointer;
    font-weight: 600;
    font-family: 'Space Grotesk', sans-serif;
    font-size: 0.95rem;
    z-index: 1000;
    transition: all 0.3s;
}

.dark-mode-toggle:hover {
    transform: scale(1.05);
    opacity: 0.9;
}

.title h1 {
    font-family: 'Cormorant Garamond', serif;
    font-size: 3rem;
    letter-spacing: 0.02em;
    margin-bottom: 0.2rem;
}

.title p {
    opacity: 0.82;
}

.panel {
    backdrop-filter: blur(8px);
    background: var(--paper);
    border: 1px solid rgba(255, 255, 255, 0.16);
    border-radius: 18px;
    padding: 0.9rem;
}

.dark-mode .panel {
    border-color: rgba(232, 228, 220, 0.22);
}

.panel label,
.panel .gr-markdown,
.panel .gradio-markdown,
.panel .gr-form label,
.panel .gr-form span {
    color: var(--ink) !important;
}

.panel textarea,
.panel input,
.panel .gr-textbox,
.panel .gr-textbox textarea,
.panel .gr-textbox input,
.panel .gr-radio,
.panel .gr-radio label,
.panel .gr-box,
.panel .gr-form {
    color: var(--ink) !important;
}

.dark-mode .panel textarea,
.dark-mode .panel input,
.dark-mode .panel .gr-textbox,
.dark-mode .panel .gr-textbox textarea,
.dark-mode .panel .gr-textbox input,
.dark-mode .panel .gr-radio,
.dark-mode .panel .gr-box,
.dark-mode .panel .gr-form {
    background: rgba(10, 10, 14, 0.9) !important;
    border-color: rgba(232, 228, 220, 0.22) !important;
}

.dark-mode .panel .gr-button,
.dark-mode .panel button {
    color: #f6f2e8 !important;
    border-color: rgba(232, 228, 220, 0.28) !important;
}

.dark-mode .panel .gr-button.gr-button-primary,
.dark-mode .panel button.primary {
    background: #3e74f2 !important;
    color: #f7f9ff !important;
}

.legend {
    display: flex;
    align-items: center;
    gap: 0.9rem;
    font-weight: 600;
    margin-bottom: 0.8rem;
}

.dot {
    display: inline-block;
    width: 10px;
    height: 10px;
    border-radius: 999px;
    margin-left: 0.7rem;
    margin-right: 0.25rem;
}

.dot.long { background: var(--long); }
.dot.short { background: var(--short); }
.dot.clear { background: var(--clear); }

.card {
    background: rgba(24, 24, 32, 0.84);
    border-radius: 14px;
    padding: 0.9rem;
    margin: 0.8rem 0;
    border: 1px solid rgba(255, 255, 255, 0.14);
    animation: rise 420ms ease both;
    color: var(--ink);
}

.dark-mode .card {
    background: rgba(14, 14, 20, 0.9);
    border: 1px solid rgba(232, 228, 220, 0.15);
}

.line-number {
    font-size: 0.8rem;
    font-weight: 700;
    text-transform: uppercase;
    letter-spacing: 0.06em;
    color: #afb0bc;
}

.dark-mode .line-number {
    color: #d3d3df;
}

.source {
    font-family: 'Cormorant Garamond', serif;
    font-size: 1.45rem;
    margin: 0.25rem 0 0.7rem;
    color: var(--source-text);
}

.chips {
    display: flex;
    flex-wrap: wrap;
    gap: 0.45rem;
}

.chip {
    display: inline-flex;
    align-items: baseline;
    gap: 0.35rem;
    border-radius: 999px;
    padding: 0.28rem 0.65rem;
    font-family: 'Cormorant Garamond', serif;
    font-size: 1.1rem;
    border: 1px solid transparent;
}

.chip small {
    font-size: 0.75rem;
    font-family: 'Space Grotesk', sans-serif;
    text-transform: uppercase;
    letter-spacing: 0.04em;
}

.chip.long {
    color: var(--chip-long-color);
    background: rgba(186, 58, 41, 0.15);
    border-color: rgba(186, 58, 41, 0.3);
}

.chip.long:before {
    content: '';
}

.dark-mode .chip.long {
    background: rgba(255, 107, 90, 0.2);
    border-color: rgba(255, 107, 90, 0.4);
}

.chip.short {
    color: var(--chip-short-color);
    background: rgba(31, 111, 109, 0.15);
    border-color: rgba(31, 111, 109, 0.3);
}

.dark-mode .chip.short {
    background: rgba(77, 217, 213, 0.2);
    border-color: rgba(77, 217, 213, 0.4);
}

.chip.clear {
    color: var(--chip-clear-color);
    background: rgba(116, 108, 95, 0.12);
    border-color: rgba(116, 108, 95, 0.25);
}

.dark-mode .chip.clear {
    color: #c8c0b0;
    background: rgba(170, 160, 144, 0.15);
    border-color: rgba(170, 160, 144, 0.3);
}

.empty {
    padding: 1rem;
    border-radius: 12px;
    background: rgba(255, 255, 255, 0.6);
    border: 1px dashed rgba(47, 43, 38, 0.2);
    color: var(--ink);
}

.dark-mode .empty {
    background: rgba(40, 35, 28, 0.7);
    border: 1px dashed rgba(232, 228, 220, 0.15);
}

@keyframes rise {
    from { transform: translateY(8px); opacity: 0; }
    to { transform: translateY(0); opacity: 1; }
}

@media (max-width: 820px) {
    .title h1 { font-size: 2.2rem; }
    .source { font-size: 1.25rem; }
    .dark-mode-toggle {
        position: relative;
        top: auto;
        right: auto;
        margin-bottom: 1rem;
    }
}
"""


with gr.Blocks() as demo:
        gr.HTML("""
        <script>
        // Detect system dark mode preference and apply on load
        function applyDarkModePreference() {
            const darkModeToggle = document.getElementById('dark-mode-toggle');
            const isDarkMode = localStorage.getItem('darkMode') === 'true' ||
                              (!localStorage.getItem('darkMode') && window.matchMedia('(prefers-color-scheme: dark)').matches);
            
            if (isDarkMode) {
                document.body.classList.add('dark-mode');
                document.documentElement.classList.add('dark-mode');
                if (darkModeToggle) darkModeToggle.textContent = '☀️ Light Mode';
            } else {
                document.body.classList.remove('dark-mode');
                document.documentElement.classList.remove('dark-mode');
                if (darkModeToggle) darkModeToggle.textContent = '🌙 Dark Mode';
            }
        }
        
        // Apply preference on page load
        window.addEventListener('load', applyDarkModePreference);
        setTimeout(applyDarkModePreference, 100);
        
        // Listen for system dark mode changes
        window.matchMedia('(prefers-color-scheme: dark)').addEventListener('change', (e) => {
            if (!localStorage.getItem('darkMode')) {
                if (e.matches) {
                    document.body.classList.add('dark-mode');
                    document.documentElement.classList.add('dark-mode');
                    document.getElementById('dark-mode-toggle').textContent = '☀️ Light Mode';
                } else {
                    document.body.classList.remove('dark-mode');
                    document.documentElement.classList.remove('dark-mode');
                    document.getElementById('dark-mode-toggle').textContent = '🌙 Dark Mode';
                }
            }
        });
        </script>
        <button id="dark-mode-toggle" class="dark-mode-toggle" onclick="
            document.body.classList.toggle('dark-mode');
            document.documentElement.classList.toggle('dark-mode');
            const isDark = document.body.classList.contains('dark-mode');
            localStorage.setItem('darkMode', isDark);
            document.getElementById('dark-mode-toggle').textContent = isDark ? '☀️ Light Mode' : '🌙 Dark Mode';
        ">🌙 Dark Mode</button>
        """)
        
        gr.Markdown(
                """
                <div class="title">
                    <h1>Ancient Greek Macronizer</h1>
                    <p>Syllable-level long/short classification with a modern, readable presentation.</p>
                </div>
                """
        )

        with gr.Column():
            with gr.Column(elem_classes=["panel"]):
                model_choice = gr.Radio(
                    label="Model",
                    choices=list(MODEL_OPTIONS.keys()),
                    value=DEFAULT_MODEL_LABEL,
                )
                text_input = gr.Textbox(
                    label="Greek Lines",
                    lines=8,
                    placeholder="Paste one or multiple lines; each line is processed separately.",
                )
                with gr.Row():
                    classify_btn = gr.Button("Classify", variant="primary")
                    clear_btn = gr.Button("Clear")
                gr.Examples(examples=examples, inputs=text_input, label="Try examples")

            with gr.Column(elem_classes=["panel"]):
                html_output = gr.HTML(label="Styled Results")
                text_output = gr.Textbox(label="Plain Output", lines=12)

        classify_btn.click(render_results, inputs=[text_input, model_choice], outputs=[html_output, text_output])
        clear_btn.click(lambda: ("", "", ""), outputs=[text_input, html_output, text_output])


if __name__ == "__main__":
    demo.launch(css=CSS)