leaderboard

README.md

@@ -1,12 +1,34 @@
----
-title: KoALa Bench
-emoji: 😻
-colorFrom: blue
-colorTo: indigo
-sdk: gradio
-sdk_version: 6.10.0
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# KoALa-Bench Gradio
+
+`KoALa-Bench`의 정적 leaderboard 페이지를 `gradio` 기반 UI로 옮긴 버전입니다.
+원본 `data/`와 `images/`도 함께 복사해 둔 상태입니다.
+
+## Run
+
+```bash
+cd /data/esseo/PycharmProject/KoALa-Bench-gradio
+pip install -r requirements.txt
+python app.py
+```
+
+If port `7860` is already in use:
+
+```bash
+GRADIO_SERVER_PORT=7861 python app.py
+```
+
+## Rebuild leaderboard data
+
+If `data/results_real` changes, rebuild `data/leaderboard-data.json` with:
+
+```bash
+cd /data/esseo/PycharmProject/KoALa-Bench-gradio
+python build_leaderboard_data.py
+```
+
+## Files
+
+- `app.py`: leaderboard 계산 로직 + Gradio UI
+- `build_leaderboard_data.py`: `results_real`를 순회해 `leaderboard-data.json` 재생성
+- `data/`: 원본 leaderboard 및 평가 결과 데이터 전체
+- `images/`: 원본 뱃지/아이콘 이미지

app.py

@@ -0,0 +1,1100 @@
+from __future__ import annotations
+
+import base64
+import json
+import os
+from functools import cmp_to_key
+from html import escape
+from pathlib import Path
+from typing import Any
+
+import gradio as gr
+
+
+HOME_VIEW = "HOME"
+TASK_ORDER = [
+    "K-disentQA",
+    "SQA",
+    "Instruct",
+    "ASR",
+    "Translation",
+    "LSQA",
+]
+
+ROOT = Path(__file__).parent
+DATA_PATH = ROOT / "data" / "leaderboard-data.json"
+IMAGES_DIR = ROOT / "images"
+
+
+def to_num(value: Any) -> float | None:
+    try:
+        num = float(value)
+    except (TypeError, ValueError):
+        return None
+    return num
+
+
+def average(values: list[float | None]) -> float | None:
+    valid = [value for value in values if value is not None]
+    if not valid:
+        return None
+    return sum(valid) / len(valid)
+
+
+def compare_scores(left: float | None, right: float | None, lower_better: bool) -> int:
+    if left is None and right is None:
+        return 0
+    if left is None:
+        return 1
+    if right is None:
+        return -1
+    if left == right:
+        return 0
+    if lower_better:
+        return -1 if left < right else 1
+    return -1 if left > right else 1
+
+
+def ordered_tasks(tasks: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    def sort_key(task: dict[str, Any]) -> tuple[int, str]:
+        try:
+            task_index = TASK_ORDER.index(task["id"])
+        except ValueError:
+            task_index = 10**6
+        return task_index, task["label"]
+
+    return sorted(tasks, key=sort_key)
+
+
+def dataset_ids(task: dict[str, Any]) -> list[str]:
+    return [dataset["id"] for dataset in task.get("datasets", [])]
+
+
+def metric_value(entry: dict[str, Any], task_id: str, dataset_id: str) -> float | None:
+    dataset = entry.get("tasks", {}).get(task_id, {}).get(dataset_id)
+    if not dataset:
+        return None
+    return to_num(dataset.get("value"))
+
+
+def metric_display(entry: dict[str, Any], task_id: str, dataset_id: str) -> str:
+    dataset = entry.get("tasks", {}).get(task_id, {}).get(dataset_id)
+    if not dataset:
+        return "-"
+    if dataset.get("display") is not None:
+        return str(dataset["display"])
+    if dataset.get("value") is None:
+        return "-"
+    return str(dataset["value"])
+
+
+def compute_task_overall(entry: dict[str, Any], task: dict[str, Any]) -> float | None:
+    return average([metric_value(entry, task["id"], dataset_id) for dataset_id in dataset_ids(task)])
+
+
+def normalize_task_scores(entries: list[dict[str, Any]], tasks: list[dict[str, Any]]) -> dict[str, dict[str, float] | None]:
+    ranges: dict[str, dict[str, float] | None] = {}
+    for task in tasks:
+        values = [entry["task_overall"][task["id"]] for entry in entries if entry["task_overall"][task["id"]] is not None]
+        if not values:
+            ranges[task["id"]] = None
+            continue
+        ranges[task["id"]] = {"min": min(values), "max": max(values)}
+    return ranges
+
+
+def normalized_score(value: float | None, score_range: dict[str, float] | None, lower_better: bool) -> float | None:
+    if value is None or score_range is None:
+        return None
+    if score_range["min"] == score_range["max"]:
+        return 100.0
+    if lower_better:
+        return ((score_range["max"] - value) / (score_range["max"] - score_range["min"])) * 100
+    return ((value - score_range["min"]) / (score_range["max"] - score_range["min"])) * 100
+
+
+def enrich_entries(entries: list[dict[str, Any]], tasks: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    entries_with_task_overall = []
+    for entry in entries:
+        task_overall = {}
+        for task in tasks:
+            task_overall[task["id"]] = compute_task_overall(entry, task)
+        entries_with_task_overall.append({**entry, "task_overall": task_overall})
+
+    task_ranges = normalize_task_scores(entries_with_task_overall, tasks)
+    enriched_entries = []
+    for entry in entries_with_task_overall:
+        normalized_task_scores = {}
+        for task in tasks:
+            normalized_task_scores[task["id"]] = normalized_score(
+                entry["task_overall"][task["id"]],
+                task_ranges[task["id"]],
+                task["lowerBetter"],
+            )
+        enriched_entries.append(
+            {
+                **entry,
+                "normalized_task_scores": normalized_task_scores,
+                "overall": average([normalized_task_scores[task["id"]] for task in tasks]),
+            }
+        )
+    return enriched_entries
+
+
+def sort_overall(entries: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    sorted_entries = sorted(
+        entries,
+        key=cmp_to_key(lambda left, right: compare_scores(left["overall"], right["overall"], False)),
+    )
+    return [{**entry, "rank": index} for index, entry in enumerate(sorted_entries, start=1)]
+
+
+def sort_task(entries: list[dict[str, Any]], task: dict[str, Any], dataset_id: str) -> list[dict[str, Any]]:
+    def compare(left: dict[str, Any], right: dict[str, Any]) -> int:
+        left_value = left["task_overall"][task["id"]] if dataset_id == "Overall" else metric_value(left, task["id"], dataset_id)
+        right_value = right["task_overall"][task["id"]] if dataset_id == "Overall" else metric_value(right, task["id"], dataset_id)
+        return compare_scores(left_value, right_value, task["lowerBetter"])
+
+    sorted_entries = sorted(entries, key=cmp_to_key(compare))
+    return [{**entry, "rank": index} for index, entry in enumerate(sorted_entries, start=1)]
+
+
+def metric_class(lower_better: bool, value: float | None) -> str:
+    if value is None:
+        return "muted"
+    return "metric-bad" if lower_better else "metric-good"
+
+
+def fmt_score(value: float | None) -> str:
+    return "-" if value is None else f"{value:.2f}"
+
+
+def image_data_uri(path: Path) -> str:
+    encoded = base64.b64encode(path.read_bytes()).decode("ascii")
+    suffix = path.suffix.lower().lstrip(".") or "png"
+    mime = "image/png" if suffix == "png" else f"image/{suffix}"
+    return f"data:{mime};base64,{encoded}"
+
+
+def load_payload() -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
+    payload = json.loads(DATA_PATH.read_text(encoding="utf-8"))
+    tasks = ordered_tasks(payload.get("tasks", []))
+    entries = enrich_entries(payload.get("entries", []), tasks)
+    return tasks, entries
+
+
+TASKS, ENTRIES = load_payload()
+TASK_MAP = {task["id"]: task for task in TASKS}
+RANKED_OVERALL = sort_overall(ENTRIES)
+BADGE_IMAGES = {
+    1: image_data_uri(IMAGES_DIR / "1st.png"),
+    2: image_data_uri(IMAGES_DIR / "2nd.png"),
+    3: image_data_uri(IMAGES_DIR / "3rd.png"),
+}
+EXTERNAL_LINK_IMAGE = image_data_uri(IMAGES_DIR / "external-link.png")
+
+
+def menu_choices() -> list[tuple[str, str]]:
+    choices = [("Home\nOverall ranking", HOME_VIEW)]
+    for task in TASKS:
+        choices.append((f"{task['label']}\n{len(task['datasets'])} datasets", task["id"]))
+    return choices
+
+
+def dataset_choices(task_id: str) -> list[str]:
+    task = TASK_MAP[task_id]
+    return ["Overall", *[dataset["id"] for dataset in task.get("datasets", [])]]
+
+
+def render_rank_strip(entries: list[dict[str, Any]]) -> str:
+    cards = []
+    for entry in entries[:12]:
+        if entry["rank"] in BADGE_IMAGES:
+            badge = f'<img class="rank-badge-image" src="{BADGE_IMAGES[entry["rank"]]}" alt="{entry["rank"]} place" />'
+        else:
+            badge = f'<span class="rank-badge">#{entry["rank"]}</span>'
+        cards.append(
+            f"""
+            <article class="rank-pill">
+              <div class="rank-badge-wrap">{badge}</div>
+              <span class="rank-name">{escape(entry["rank_name"])}</span>
+            </article>
+            """
+        )
+    return f"""
+    <section class="section-card card">
+      <div class="section-head">
+        <h3>Top Ranking</h3>
+      </div>
+      <div class="rank-strip-list">
+        {"".join(cards)}
+      </div>
+    </section>
+    """
+
+
+def render_home_table(entries: list[dict[str, Any]]) -> str:
+    header_top = [
+        '<th rowspan="2" class="rank-col col-rank">Rank</th>',
+        '<th rowspan="2" class="col-rankname">RankName</th>',
+        '<th rowspan="2" class="col-model">Model</th>',
+        '<th rowspan="2">URL</th>',
+        '<th rowspan="2">Overall</th>',
+        *[f'<th class="grouped" colspan="1">{escape(task["label"])}</th>' for task in TASKS],
+    ]
+    header_bottom = [f'<th>{escape(task["shortMetric"])}</th>' for task in TASKS]
+
+    rows = []
+    for entry in entries:
+        url = entry.get("url") or ""
+        if url:
+            url_cell = (
+                f'<a class="url-link" href="{escape(url)}" target="_blank" rel="noopener noreferrer" '
+                f'aria-label="External link"><img src="{EXTERNAL_LINK_IMAGE}" alt="" /></a>'
+            )
+        else:
+            url_cell = "-"
+
+        task_cells = []
+        for task in TASKS:
+            value = entry["task_overall"][task["id"]]
+            task_cells.append(f'<td><span class="{metric_class(task["lowerBetter"], value)}">{fmt_score(value)}</span></td>')
+
+        rows.append(
+            "<tr>"
+            f'<td class="rank-col col-rank">{entry["rank"]}</td>'
+            f'<td class="col-rankname">{escape(entry["rank_name"])}</td>'
+            f'<td class="col-model">{escape(entry.get("model") or entry["rank_name"])}</td>'
+            f"<td>{url_cell}</td>"
+            f"<td>{fmt_score(entry['overall'])}</td>"
+            f"{''.join(task_cells)}"
+            "</tr>"
+        )
+
+    return f"""
+    <section class="section-card card">
+      <div class="section-head">
+        <h3>Overall Leaderboard</h3>
+      </div>
+      <div class="table-scroll">
+        <table>
+          <colgroup>
+            <col class="col-rank" />
+            <col class="col-rankname" />
+            <col class="col-model" />
+            <col />
+            <col />
+            {"".join("<col />" for _ in TASKS)}
+          </colgroup>
+          <thead>
+            <tr>{"".join(header_top)}</tr>
+            <tr>{"".join(header_bottom)}</tr>
+          </thead>
+          <tbody>{"".join(rows)}</tbody>
+        </table>
+      </div>
+    </section>
+    """
+
+
+def render_home() -> str:
+    return f"{render_rank_strip(RANKED_OVERALL)}{render_home_table(RANKED_OVERALL)}"
+
+
+def render_task_title(task: dict[str, Any]) -> str:
+    return f"""
+    <div class="section-head">
+      <div>
+        <h3 class="task-title">Task : {escape(task["label"])}</h3>
+      </div>
+    </div>
+    """
+
+
+def render_task_table(task: dict[str, Any], dataset_id: str) -> str:
+    ranked_entries = sort_task(ENTRIES, task, dataset_id)
+    active_label = "Overall"
+    if dataset_id != "Overall":
+        active_label = next(
+            (dataset["label"] for dataset in task["datasets"] if dataset["id"] == dataset_id),
+            dataset_id,
+        )
+
+    rows = []
+    for entry in ranked_entries:
+        numeric_value = entry["task_overall"][task["id"]] if dataset_id == "Overall" else metric_value(entry, task["id"], dataset_id)
+        display_value = fmt_score(numeric_value) if dataset_id == "Overall" else metric_display(entry, task["id"], dataset_id)
+        rows.append(
+            "<tr>"
+            f'<td class="rank-col col-rank">{entry["rank"]}</td>'
+            f'<td class="col-rankname">{escape(entry["rank_name"])}</td>'
+            f'<td class="col-model">{escape(entry.get("model") or entry["rank_name"])}</td>'
+            f'<td><span class="{metric_class(task["lowerBetter"], numeric_value)}">{escape(display_value)}</span></td>'
+            "</tr>"
+        )
+
+    return f"""
+    <div id="taskTableMount">
+      <div class="table-scroll">
+        <table class="task-performance-table">
+          <colgroup>
+            <col class="col-rank" />
+            <col class="col-rankname" />
+            <col class="col-model" />
+            <col />
+          </colgroup>
+          <thead>
+            <tr>
+              <th class="rank-col col-rank">Rank</th>
+              <th class="col-rankname">RankName</th>
+              <th class="col-model">Model</th>
+              <th>{escape(active_label)}</th>
+            </tr>
+          </thead>
+          <tbody>{"".join(rows)}</tbody>
+        </table>
+      </div>
+    </div>
+    """
+
+
+def update_view(active_view: str, current_dataset: str | None) -> tuple[Any, Any, str, str, str]:
+    if active_view == HOME_VIEW:
+        first_task = TASKS[0]
+        return (
+            gr.update(visible=True),
+            gr.update(visible="hidden"),
+            gr.update(choices=dataset_choices(first_task["id"]), value="Overall"),
+            render_task_title(first_task),
+            first_task["metricLabel"],
+            render_task_table(first_task, "Overall"),
+        )
+
+    task = TASK_MAP[active_view]
+    choices = dataset_choices(active_view)
+    dataset_id = current_dataset if current_dataset in choices else "Overall"
+    return (
+        gr.update(visible="hidden"),
+        gr.update(visible=True),
+        gr.update(choices=choices, value=dataset_id),
+        render_task_title(task),
+        task["metricLabel"],
+        render_task_table(task, dataset_id),
+    )
+
+
+CUSTOM_CSS = """
+@import url('https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@400;500;700&family=Noto+Sans+KR:wght@400;500;700&display=swap');
+
+:root {
+  --bg: #f2f4f8;
+  --bg-strong: #dde5f2;
+  --panel: rgba(255, 255, 255, 0.84);
+  --panel-strong: #ffffff;
+  --text: #0c1730;
+  --muted: #66748b;
+  --line: rgba(12, 23, 48, 0.14);
+  --primary: #0e3a8a;
+  --accent: #c56b12;
+  --pending: #74674a;
+  --pending-bg: #f4ecd8;
+  --success: #0c8f61;
+  --danger: #b8612f;
+  --shadow: 0 14px 38px rgba(12, 23, 48, 0.08);
+}
+
+html, body, .gradio-container {
+  margin: 0 !important;
+  min-height: 100vh;
+  font-family: "Noto Sans KR", "Space Grotesk", sans-serif !important;
+  color: var(--text);
+  background:
+    radial-gradient(circle at 15% 0%, #dce8ff 0%, transparent 32%),
+    radial-gradient(circle at 95% 5%, #ffe4cf 0%, transparent 24%),
+    linear-gradient(180deg, #f8fbff 0%, #eef2f7 100%) !important;
+}
+
+.gradio-container {
+  max-width: 100% !important;
+}
+
+.app-root {
+  position: relative;
+  max-width: 1440px;
+  margin: 0 auto;
+  padding: 24px 20px 56px;
+  overflow-x: auto;
+  -webkit-overflow-scrolling: touch;
+}
+
+.bg-orb {
+  position: fixed;
+  border-radius: 999px;
+  filter: blur(70px);
+  opacity: 0.45;
+  pointer-events: none;
+  z-index: 0;
+}
+
+.orb-1 {
+  width: 280px;
+  height: 280px;
+  background: #a7c5ff;
+  top: -110px;
+  left: -80px;
+}
+
+.orb-2 {
+  width: 240px;
+  height: 240px;
+  background: #ffd2a4;
+  top: -70px;
+  right: -60px;
+}
+
+.layout-row {
+  position: relative;
+  z-index: 1;
+  flex-wrap: nowrap !important;
+  gap: 24px;
+  align-items: flex-start;
+  min-width: 1180px;
+}
+
+.sidebar-panel {
+  flex: 0 0 260px !important;
+  width: 260px !important;
+  min-width: 260px !important;
+  max-width: 260px !important;
+  position: sticky;
+  top: 20px;
+  align-self: start;
+  border: 1px solid var(--line);
+  border-radius: 26px;
+  background: rgba(7, 17, 40, 0.95);
+  color: #f5f7fb;
+  padding: 22px 18px;
+  box-shadow: var(--shadow);
+}
+
+.sidebar-head {
+  margin-bottom: 26px;
+}
+
+.sidebar-kicker, .kicker {
+  margin: 0;
+  letter-spacing: 0.12em;
+  text-transform: uppercase;
+  font-size: 12px;
+}
+
+.sidebar-head h1 {
+  margin: 8px 0 0;
+  font-size: 34px;
+  line-height: 0.95;
+  color: #f5f7fb;
+}
+
+.content-panel {
+  flex: 1 1 auto !important;
+  min-width: 0;
+}
+
+.hero {
+  margin-bottom: 20px;
+}
+
+.hero-topline {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 20px;
+  text-align: center;
+}
+
+.hero-topline > div {
+  flex: 1;
+}
+
+.hero h2 {
+  margin: 8px 0 6px;
+  font-family: "Space Grotesk", sans-serif;
+  font-size: clamp(42px, 7vw, 76px);
+  line-height: 0.95;
+  color: var(--text);
+}
+
+.desc {
+  margin: 0 auto;
+  max-width: 760px;
+  color: var(--muted);
+  font-size: 17px;
+}
+
+.card {
+  background: var(--panel);
+  border: 1px solid var(--line);
+  border-radius: 24px;
+  box-shadow: var(--shadow);
+  backdrop-filter: blur(8px);
+}
+
+.section-card {
+  padding: 18px;
+}
+
+.section-head {
+  display: flex;
+  align-items: baseline;
+  justify-content: space-between;
+  gap: 12px;
+  margin-bottom: 12px;
+}
+
+.section-head h3,
+.task-title {
+  margin: 0;
+  font-size: 28px;
+  font-family: "Space Grotesk", sans-serif;
+}
+
+.rank-strip-list {
+  display: grid;
+  grid-template-columns: repeat(6, minmax(120px, 1fr));
+  gap: 10px;
+}
+
+.rank-pill {
+  border: 1px solid var(--line);
+  border-radius: 16px;
+  padding: 12px;
+  background: linear-gradient(135deg, #ffffff 0%, var(--bg-strong) 100%);
+}
+
+.rank-badge-wrap {
+  min-height: 28px;
+}
+
+.rank-badge {
+  display: inline-block;
+  padding: 3px 8px;
+  border-radius: 999px;
+  background: var(--primary);
+  color: #fff;
+  font-size: 12px;
+}
+
+.rank-badge-image {
+  display: block;
+  width: auto;
+  height: 28px;
+}
+
+.rank-name {
+  display: block;
+  margin-top: 8px;
+  font-weight: 700;
+}
+
+.table-scroll {
+  overflow-x: auto;
+  overflow-y: visible;
+  border-radius: 0;
+  border: 1px solid #ffffff !important;
+  background: var(--panel-strong);
+}
+
+table {
+  width: 100%;
+  min-width: 1080px;
+  border-collapse: collapse;
+  border: 1px solid #ffffff !important;
+}
+
+.task-performance-table {
+  table-layout: fixed;
+}
+
+thead th {
+  background: #e8edf6;
+  border-bottom: 1px solid #ffffff !important;
+  white-space: nowrap;
+}
+
+thead tr:first-child th.grouped {
+  text-align: center;
+}
+
+th, td {
+  padding: 12px 14px;
+  text-align: left;
+  border-bottom: 1px solid #ffffff !important;
+  border-right: 1px solid #ffffff !important;
+  font-size: 14px;
+}
+
+th:first-child,
+td:first-child {
+  border-left: 1px solid #ffffff !important;
+}
+
+th:last-child,
+td:last-child {
+  border-right: 0;
+}
+
+.table-scroll table,
+.table-scroll thead,
+.table-scroll tbody,
+.table-scroll tr,
+.table-scroll th,
+.table-scroll td {
+  border-color: #ffffff !important;
+}
+
+tbody tr:hover {
+  background: #f8fbff;
+}
+
+.rank-col {
+  font-family: "Space Grotesk", sans-serif;
+  font-weight: 700;
+  width: 72px;
+}
+
+.col-rank {
+  width: 96px;
+}
+
+.col-rankname {
+  width: 240px;
+}
+
+.col-model {
+  width: 320px;
+}
+
+.url-link {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  width: 24px;
+  height: 24px;
+}
+
+.url-link img {
+  display: block;
+  width: 18px;
+  height: 18px;
+}
+
+.metric-good {
+  color: var(--success);
+  font-weight: 700;
+}
+
+.metric-bad {
+  color: var(--danger);
+  font-weight: 700;
+}
+
+.muted {
+  color: var(--muted);
+}
+
+.task-filters {
+  display: flex;
+  flex-wrap: nowrap;
+  align-items: flex-start;
+  gap: 12px;
+  margin-top: 12px;
+  width: 100%;
+  overflow-x: auto;
+  -webkit-overflow-scrolling: touch;
+  padding-bottom: 6px;
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+}
+
+.task-filters .dataset-wrap { flex: 0 0 620px; min-width: 620px; }
+.task-filters .metric-wrap { flex: 0 0 340px; min-width: 340px; }
+
+.task-filters .dataset-wrap,
+.task-filters .metric-wrap,
+.task-filters .dataset-wrap > div,
+.task-filters .metric-wrap > div {
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+}
+
+/* Keep Dataset / Metric columns pinned to the same top baseline. */
+.task-filters .dataset-wrap,
+.task-filters .metric-wrap {
+  align-self: flex-start !important;
+  justify-self: flex-start !important;
+  margin-top: 0 !important;
+  padding-top: 0 !important;
+}
+
+.task-view-shell .filter-title {
+  margin: 0 0 6px 0 !important;
+  font-size: 13px !important;
+  color: var(--muted) !important;
+  line-height: 1.2 !important;
+}
+
+#taskTableMount {
+  margin-top: 18px;
+}
+
+.task-menu-radio {
+  gap: 8px;
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+  padding: 0 !important;
+}
+
+.task-menu-radio > div,
+.task-menu-radio .block,
+.task-menu-radio .gradio-radio,
+.task-menu-radio .form {
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+  padding: 0 !important;
+}
+
+.task-menu-radio label > span,
+.task-menu-radio label > div,
+.task-menu-radio label .wrap {
+  white-space: pre-line !important;
+}
+
+.task-menu-radio label:has(input[type="radio"]) {
+  width: 100%;
+  margin: 0 !important;
+  border: 1px solid rgba(255, 255, 255, 0.1) !important;
+  background: rgba(255, 255, 255, 0.05) !important;
+  color: #f5f7fb !important;
+  border-radius: 14px !important;
+  padding: 12px 14px !important;
+  cursor: pointer !important;
+  min-height: 72px;
+  align-content: center;
+  box-shadow: none !important;
+  transition: background 0.18s ease, color 0.18s ease, border-color 0.18s ease;
+}
+
+.task-menu-radio label:has(input[type="radio"]):hover {
+  background: rgba(255, 255, 255, 0.1) !important;
+}
+
+.task-menu-radio label:has(input[type="radio"]:checked) {
+  background: linear-gradient(135deg, #fcf7eb 0%, #dfeaff 100%) !important;
+  color: var(--text) !important;
+  border-color: transparent !important;
+}
+
+.task-menu-radio input[type="radio"] {
+  display: none !important;
+}
+
+.task-menu-radio label span,
+.task-menu-radio label div,
+.task-menu-radio label p {
+  color: inherit !important;
+}
+
+.task-menu-radio label:has(input[type="radio"]) span,
+.task-menu-radio label:has(input[type="radio"]) div {
+  color: #f5f7fb !important;
+}
+
+.task-menu-radio label:has(input[type="radio"]:checked) span,
+.task-menu-radio label:has(input[type="radio"]:checked) div {
+  color: var(--text) !important;
+}
+
+.task-menu-radio .wrap,
+.task-menu-radio label span:last-child {
+  opacity: 0.72;
+  font-size: 12px;
+}
+
+.task-view-shell .gradio-radio,
+.task-view-shell .gradio-textbox {
+  margin: 0 !important;
+  min-width: 0 !important;
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+}
+
+.task-view-shell .gradio-radio label,
+.task-view-shell .gradio-textbox label {
+  font-size: 13px !important;
+  color: var(--muted) !important;
+}
+
+.task-view-shell .dataset-radio {
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+  padding: 0 !important;
+  margin-left: 0 !important;
+  margin-bottom: 12px !important;
+  max-width: 100% !important;
+  width: 100% !important;
+}
+
+.task-view-shell .dataset-radio > div,
+.task-view-shell .dataset-radio .block,
+.task-view-shell .dataset-radio .form {
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+  padding: 0 !important;
+  display: grid !important;
+  grid-template-columns: repeat(2, minmax(180px, 1fr)) !important;
+  gap: 8px 10px !important;
+  align-items: start !important;
+  justify-content: start !important;
+  min-height: 86px !important;
+  align-content: start !important;
+  max-width: 100% !important;
+  width: 100% !important;
+}
+
+.task-view-shell .dataset-radio label:has(input[type="radio"]) {
+  margin: 0 !important;
+  border: 1px solid var(--line) !important;
+  background: rgba(255, 255, 255, 0.88) !important;
+  color: var(--text) !important;
+  border-radius: 10px !important;
+  padding: 0 12px !important;
+  height: 40px !important;
+  min-height: 40px !important;
+  width: auto !important;
+  display: flex !important;
+  align-items: center !important;
+  box-shadow: none !important;
+  justify-content: flex-start !important;
+  font-size: 14px !important;
+  line-height: 1.2 !important;
+}
+
+.task-view-shell .dataset-radio label:has(input[type="radio"]:checked) {
+  background: linear-gradient(135deg, #fcf7eb 0%, #dfeaff 100%) !important;
+  border-color: rgba(14, 58, 138, 0.22) !important;
+  font-weight: 700 !important;
+}
+
+.task-view-shell .dataset-radio input[type="radio"] {
+  display: none !important;
+}
+
+.task-view-shell .metric-field,
+.task-view-shell .dataset-field {
+  align-self: flex-start !important;
+  background: transparent !important;
+}
+
+/* Remove any residual top spacing on metric box so it aligns with Dataset. */
+.task-view-shell .metric-wrap .metric-field,
+.task-view-shell .metric-wrap .gradio-textbox,
+.task-view-shell .metric-wrap .gradio-textbox > div,
+.task-view-shell .metric-wrap .gradio-textbox .block,
+.task-view-shell .metric-wrap .gradio-textbox .form,
+.task-view-shell .metric-wrap .gradio-textbox .wrap {
+  margin-top: 0 !important;
+  padding-top: 0 !important;
+}
+
+.task-view-shell .metric-field > div,
+.task-view-shell .dataset-field > div {
+  background: transparent !important;
+}
+
+.task-view-shell .metric-field .gradio-textbox,
+.task-view-shell .metric-field .gradio-textbox > div,
+.task-view-shell .metric-field .gradio-textbox .block,
+.task-view-shell .metric-field .gradio-textbox .form {
+  background: transparent !important;
+  border: 0 !important;
+  box-shadow: none !important;
+}
+
+.task-view-shell .metric-field,
+.task-view-shell .metric-field .gradio-textbox,
+.task-view-shell .metric-field .gradio-textbox .wrap {
+  width: 100% !important;
+  max-width: none !important;
+}
+
+.task-view-shell .metric-field .gradio-textbox .wrap,
+.task-view-shell .metric-field .gradio-textbox textarea,
+.task-view-shell .metric-field .gradio-textbox input {
+  height: 40px !important;
+  min-height: 40px !important;
+  width: 100% !important;
+}
+
+.task-view-shell input,
+.task-view-shell textarea,
+.task-view-shell .wrap-inner,
+.task-view-shell button.secondary-down-arrow,
+.task-view-shell .gradio-textbox .wrap {
+  border-radius: 12px !important;
+}
+
+.task-view-shell .gradio-textbox .wrap,
+.task-view-shell .gradio-textbox textarea,
+.task-view-shell .gradio-textbox input {
+  border: 1px solid var(--line) !important;
+  background: rgba(255, 255, 255, 0.88) !important;
+  color: var(--text) !important;
+}
+
+@media (max-width: 1280px) {
+  .layout-row {
+    min-width: 1120px;
+  }
+
+  .task-filters .dataset-wrap { flex-basis: 560px; min-width: 560px; }
+  .task-filters .metric-wrap { flex-basis: 320px; min-width: 320px; }
+
+  .task-view-shell .dataset-radio > div,
+  .task-view-shell .dataset-radio .block,
+  .task-view-shell .dataset-radio .form {
+    grid-template-columns: repeat(2, minmax(150px, 1fr)) !important;
+  }
+}
+
+@media (max-width: 980px) {
+  .layout-row {
+    min-width: 1040px;
+  }
+
+  .rank-strip-list {
+    grid-template-columns: repeat(2, minmax(0, 1fr));
+  }
+
+  .task-view-shell .dataset-radio > div,
+  .task-view-shell .dataset-radio .block,
+  .task-view-shell .dataset-radio .form {
+    grid-template-columns: repeat(2, minmax(140px, 1fr)) !important;
+  }
+}
+
+@media (max-width: 720px) {
+  .app-root {
+    padding: 16px 14px 40px;
+  }
+
+  .hero-topline {
+    flex-direction: column;
+    align-items: center;
+  }
+
+  .hero h2 {
+    font-size: 40px;
+  }
+
+  .rank-strip-list {
+    grid-template-columns: 1fr;
+  }
+
+  .task-view-shell .dataset-radio > div,
+  .task-view-shell .dataset-radio .block,
+  .task-view-shell .dataset-radio .form {
+    grid-template-columns: 1fr !important;
+  }
+}
+"""
+
+
+def build_app() -> gr.Blocks:
+    with gr.Blocks(title="Ko-Speech-Eval Leaderboard", fill_width=True) as demo:
+        with gr.Column(elem_classes=["app-root"]):
+            gr.HTML('<div class="bg-orb orb-1"></div><div class="bg-orb orb-2"></div>')
+            with gr.Row(elem_classes=["layout-row"]):
+                with gr.Column(scale=0, min_width=260, elem_classes=["sidebar-panel"]):
+                    gr.HTML(
+                        """
+                        <div class="sidebar-head">
+                          <p class="sidebar-kicker">KoALa-bench</p>
+                          <h1>Leaderboard</h1>
+                        </div>
+                        """
+                    )
+                    menu = gr.Radio(
+                        choices=menu_choices(),
+                        value=HOME_VIEW,
+                        show_label=False,
+                        container=False,
+                        elem_classes=["task-menu", "task-menu-radio"],
+                    )
+
+                with gr.Column(scale=1, elem_classes=["content-panel"]):
+                    gr.HTML(
+                        """
+                        <header class="hero">
+                          <div class="hero-topline">
+                            <div>
+                              <p class="kicker">Korean Audio Language benchmark</p>
+                              <h2>Leaderboard for KoALa</h2>
+                            </div>
+                          </div>
+                        </header>
+                        """
+                    )
+
+                    home_view = gr.HTML(render_home(), visible=True)
+
+                    with gr.Column(visible="hidden", elem_classes=["task-view-shell", "section-card", "card"]) as task_view:
+                        task_title = gr.HTML()
+                        with gr.Row(elem_classes=["task-filters"]):
+                            with gr.Column(scale=3, min_width=420, elem_classes=["dataset-wrap"]):
+                                gr.HTML('<p class="filter-title">Dataset</p>')
+                                dataset_dropdown = gr.Radio(
+                                    choices=dataset_choices(TASKS[0]["id"]),
+                                    value="Overall",
+                                    show_label=False,
+                                    elem_classes=["dataset-field", "dataset-radio"],
+                                )
+                            with gr.Column(scale=2, min_width=280, elem_classes=["metric-wrap"]):
+                                gr.HTML('<p class="filter-title">Metric</p>')
+                                metric_text = gr.Textbox(
+                                    show_label=False,
+                                    interactive=False,
+                                    elem_classes=["metric-field"],
+                                )
+                        task_table = gr.HTML()
+
+        menu.change(
+            fn=update_view,
+            inputs=[menu, dataset_dropdown],
+            outputs=[home_view, task_view, dataset_dropdown, task_title, metric_text, task_table],
+        )
+        dataset_dropdown.change(
+            fn=update_view,
+            inputs=[menu, dataset_dropdown],
+            outputs=[home_view, task_view, dataset_dropdown, task_title, metric_text, task_table],
+        )
+        demo.load(
+            fn=update_view,
+            inputs=[menu, dataset_dropdown],
+            outputs=[home_view, task_view, dataset_dropdown, task_title, metric_text, task_table],
+        )
+    return demo
+
+
+if __name__ == "__main__":
+    app = build_app()
+    app.launch(
+        server_port=int(os.getenv("GRADIO_SERVER_PORT", "7860")),
+        css=CUSTOM_CSS,
+    )

build_leaderboard_data.py

@@ -0,0 +1,292 @@
+from __future__ import annotations
+
+import json
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+
+ROOT = Path(__file__).parent
+RESULTS_ROOT = ROOT / "data" / "results_real"
+LEADERBOARD_JSON = ROOT / "data" / "leaderboard-data.json"
+
+
+CANONICAL_TASKS = [
+    {
+        "id": "K-disentQA",
+        "label": "SCA-QA",
+        "metricLabel": "Speech Context Faithfulness",
+        "shortMetric": "Faithfulness",
+        "lowerBetter": False,
+        "datasets": [
+            {"id": "history_after_chosun", "label": "History_after_chosun"},
+            {"id": "history_after_chosun_other", "label": "History_after_chosun Other"},
+            {"id": "history_before_chosun", "label": "History_before_chosun"},
+            {"id": "history_before_chosun_other", "label": "History_before_chosun Other"},
+            {"id": "k-sports", "label": "K-sports"},
+            {"id": "k-sports_other", "label": "K-sports Other"},
+            {"id": "kpop", "label": "K-pop"},
+            {"id": "kpop_other", "label": "K-pop Other"},
+        ],
+    },
+    {
+        "id": "SQA",
+        "label": "Speech QA",
+        "metricLabel": "Accuracy (%)",
+        "shortMetric": "Acc(%)",
+        "lowerBetter": False,
+        "datasets": [
+            {"id": "click", "label": "CLICk"},
+            {"id": "click_other", "label": "CLICk Other"},
+            {"id": "kobest_boolq", "label": "KoBest BoolQ"},
+            {"id": "kobest_boolq_other", "label": "KoBest BoolQ Other"},
+        ],
+    },
+    {
+        "id": "Instruct",
+        "label": "Speech Instruction",
+        "metricLabel": "Score (GPT-4o as Judge)",
+        "shortMetric": "Score (GPT-4o as Judge)",
+        "lowerBetter": False,
+        "datasets": [
+            {"id": "alpaca", "label": "Alpaca"},
+            {"id": "alpaca_other", "label": "Alpaca Other"},
+            {"id": "kudge", "label": "KUDGE"},
+            {"id": "kudge_other", "label": "KUDGE Other"},
+            {"id": "openhermes", "label": "OpenHermes"},
+            {"id": "openhermes_other", "label": "OpenHermes Other"},
+            {"id": "vicuna", "label": "Vicuna"},
+            {"id": "vicuna_other", "label": "Vicuna Other"},
+        ],
+    },
+    {
+        "id": "ASR",
+        "label": "ASR",
+        "metricLabel": "CER (%)",
+        "shortMetric": "CER",
+        "lowerBetter": True,
+        "datasets": [
+            {"id": "common_voice_korea", "label": "CommonVoice-KO"},
+            {"id": "common_voice_korea_other", "label": "CommonVoice-KO Other"},
+            {"id": "ksponspeech_eval_clean", "label": "KsponSpeech Clean"},
+            {"id": "ksponspeech_eval_other", "label": "KsponSpeech Other"},
+            {"id": "zeroth_korean_test", "label": "Zeroth-Korean"},
+            {"id": "zeroth_korean_test_other", "label": "Zeroth-Korean Other"},
+        ],
+    },
+    {
+        "id": "Translation",
+        "label": "Translation",
+        "metricLabel": "BLEU / METEOR",
+        "shortMetric": "BLEU / METEOR",
+        "lowerBetter": False,
+        "datasets": [
+            {"id": "etri_tst-COMMON", "label": "ETRI-TST-Common"},
+            {"id": "etri_tst-HE", "label": "ETRI-TST-HE"},
+        ],
+    },
+    {
+        "id": "LSQA",
+        "label": "Long Speech Understanding",
+        "metricLabel": "Accuracy (%)",
+        "shortMetric": "Acc(%)",
+        "lowerBetter": False,
+        "datasets": [
+            {"id": "mctest", "label": "MCTest"},
+            {"id": "mctest_other", "label": "MCTest Other"},
+        ],
+    },
+]
+
+
+FOLDER_TO_DATASET_ID = {
+    "K-disentQA": {
+        "history_after_chosun": "history_after_chosun",
+        "history_after_chosun_other": "history_after_chosun_other",
+        "history_before_chosun": "history_before_chosun",
+        "history_before_chosun_other": "history_before_chosun_other",
+        "k-sports": "k-sports",
+        "k-sports_other": "k-sports_other",
+        "kpop": "kpop",
+        "kpop_other": "kpop_other",
+    },
+    "SQA": {
+        "click": "click",
+        "click_other": "click_other",
+        "kobest_boolq": "kobest_boolq",
+        "kobest_boolq_other": "kobest_boolq_other",
+    },
+    "Instruct": {
+        "alpaca": "alpaca",
+        "alpaca_other": "alpaca_other",
+        "kudge": "kudge",
+        "kudge_other": "kudge_other",
+        "openhermes": "openhermes",
+        "openhermes_other": "openhermes_other",
+        "vicuna": "vicuna",
+        "vicuna_other": "vicuna_other",
+    },
+    "ASR": {
+        "common_voice_korea": "common_voice_korea",
+        "common_voice_korea_other": "common_voice_korea_other",
+        "ksponspeech_eval_clean": "ksponspeech_eval_clean",
+        "ksponspeech_eval_other": "ksponspeech_eval_other",
+        "zeroth_korean_test": "zeroth_korean_test",
+        "zeroth_korean_test_other": "zeroth_korean_test_other",
+    },
+    "Translation": {
+        "etri_tst-COMMON": "etri_tst-COMMON",
+        "etri_tst-HE": "etri_tst-HE",
+    },
+    "LSQA": {
+        "mctest": "mctest",
+        "mctest_other": "mctest_other",
+    },
+}
+
+
+def load_existing_entry_meta() -> dict[str, dict[str, str]]:
+    if not LEADERBOARD_JSON.exists():
+        return {}
+
+    payload = json.loads(LEADERBOARD_JSON.read_text(encoding="utf-8"))
+    return {
+        entry["id"]: {
+            "rank_name": entry.get("rank_name", entry["id"]),
+            "model": entry.get("model", ""),
+            "url": entry.get("url", ""),
+        }
+        for entry in payload.get("entries", [])
+    }
+
+
+def pick_summary(dataset_dir: Path) -> Path | None:
+    direct = sorted(path for path in dataset_dir.glob("*_summary.json") if path.is_file())
+    if direct:
+        return direct[0]
+
+    recursive = sorted(
+        dataset_dir.rglob("*_summary.json"),
+        key=lambda path: (len(path.relative_to(dataset_dir).parts), str(path)),
+    )
+    return recursive[0] if recursive else None
+
+
+def extract_metric(task_name: str, payload: dict[str, Any]) -> dict[str, Any] | None:
+    if task_name == "K-disentQA":
+        value = payload.get("accuracy_speech")
+        if value is None:
+            return None
+        value *= 100
+        return {"value": value, "display": f"{value:.2f}"}
+
+    if task_name in {"SQA", "LSQA"}:
+        value = payload.get("accuracy_logit")
+        if value is None:
+            value = payload.get("accuracy_generation")
+        if value is None:
+            return None
+        value *= 100
+        return {"value": value, "display": f"{value:.2f}"}
+
+    if task_name == "Instruct":
+        value = payload.get("avg_gpt_score")
+        if value is None:
+            return None
+        value *= 100
+        return {"value": value, "display": f"{value:.2f}"}
+
+    if task_name == "ASR":
+        value = payload.get("total_cer")
+        if value is None:
+            return None
+        value *= 100
+        return {"value": value, "display": f"{value:.2f}"}
+
+    if task_name == "Translation":
+        bleu = payload.get("avg_bleu")
+        if bleu is None:
+            bleu = payload.get("corpus_bleu")
+        meteor = payload.get("avg_meteor")
+        if bleu is None:
+            return None
+        if meteor is None:
+            return {"value": bleu, "display": f"{bleu:.2f}"}
+        return {"value": bleu, "display": f"{bleu:.2f} / {meteor:.2f}"}
+
+    return None
+
+
+def build_leaderboard_payload() -> dict[str, Any]:
+    if not RESULTS_ROOT.exists():
+        raise SystemExit(f"Missing results directory: {RESULTS_ROOT}")
+
+    existing_meta = load_existing_entry_meta()
+    entries: dict[str, dict[str, Any]] = {}
+
+    for task in CANONICAL_TASKS:
+        task_id = task["id"]
+        task_dir = RESULTS_ROOT / task_id
+        if not task_dir.exists():
+            continue
+
+        folder_map = FOLDER_TO_DATASET_ID[task_id]
+        for model_dir in sorted(path for path in task_dir.iterdir() if path.is_dir()):
+            model_id = model_dir.name
+            meta = existing_meta.get(model_id, {})
+            entry = entries.setdefault(
+                model_id,
+                {
+                    "id": model_id,
+                    "rank_name": meta.get("rank_name", model_id),
+                    "model": meta.get("model", ""),
+                    "url": meta.get("url", ""),
+                    "tasks": {},
+                },
+            )
+            entry["tasks"].setdefault(task_id, {})
+
+            for dataset_dir in sorted(path for path in model_dir.iterdir() if path.is_dir()):
+                dataset_id = folder_map.get(dataset_dir.name)
+                if not dataset_id:
+                    continue
+
+                summary_path = pick_summary(dataset_dir)
+                if summary_path is None:
+                    continue
+
+                payload = json.loads(summary_path.read_text(encoding="utf-8"))
+                metric = extract_metric(task_id, payload)
+                if metric is None:
+                    continue
+
+                if not entry["model"] and payload.get("model"):
+                    entry["model"] = payload["model"]
+                entry["tasks"][task_id][dataset_id] = metric
+
+    for entry in entries.values():
+        if not entry["model"]:
+            entry["model"] = entry["id"]
+        for task in CANONICAL_TASKS:
+            entry["tasks"].setdefault(task["id"], {})
+
+    return {
+        "generatedAt": datetime.now(timezone.utc).isoformat(),
+        "sourceRoot": "data/results_real",
+        "tasks": CANONICAL_TASKS,
+        "entries": [entries[entry_id] for entry_id in sorted(entries)],
+    }
+
+
+def main() -> None:
+    payload = build_leaderboard_payload()
+    LEADERBOARD_JSON.write_text(
+        json.dumps(payload, ensure_ascii=False, indent=2) + "\n",
+        encoding="utf-8",
+    )
+    print(f"Wrote {LEADERBOARD_JSON}")
+    print(f"Entries: {len(payload['entries'])}")
+
+
+if __name__ == "__main__":
+    main()

data/leaderboard-data.json

@@ -0,0 +1,1022 @@
+{
+  "generatedAt": "2026-03-30T07:45:11.033924+00:00",
+  "sourceRoot": "data/results_real",
+  "tasks": [
+    {
+      "id": "K-disentQA",
+      "label": "SCA-QA",
+      "metricLabel": "Speech Context Faithfulness",
+      "shortMetric": "Faithfulness",
+      "lowerBetter": false,
+      "datasets": [
+        {
+          "id": "history_after_chosun",
+          "label": "History_after_chosun"
+        },
+        {
+          "id": "history_after_chosun_other",
+          "label": "History_after_chosun Other"
+        },
+        {
+          "id": "history_before_chosun",
+          "label": "History_before_chosun"
+        },
+        {
+          "id": "history_before_chosun_other",
+          "label": "History_before_chosun Other"
+        },
+        {
+          "id": "k-sports",
+          "label": "K-sports"
+        },
+        {
+          "id": "k-sports_other",
+          "label": "K-sports Other"
+        },
+        {
+          "id": "kpop",
+          "label": "K-pop"
+        },
+        {
+          "id": "kpop_other",
+          "label": "K-pop Other"
+        }
+      ]
+    },
+    {
+      "id": "SQA",
+      "label": "Speech QA",
+      "metricLabel": "Accuracy (%)",
+      "shortMetric": "Acc(%)",
+      "lowerBetter": false,
+      "datasets": [
+        {
+          "id": "click",
+          "label": "CLICk"
+        },
+        {
+          "id": "click_other",
+          "label": "CLICk Other"
+        },
+        {
+          "id": "kobest_boolq",
+          "label": "KoBest BoolQ"
+        },
+        {
+          "id": "kobest_boolq_other",
+          "label": "KoBest BoolQ Other"
+        }
+      ]
+    },
+    {
+      "id": "Instruct",
+      "label": "Speech Instruction",
+      "metricLabel": "Score (GPT-4o as Judge)",
+      "shortMetric": "Score (GPT-4o as Judge)",
+      "lowerBetter": false,
+      "datasets": [
+        {
+          "id": "alpaca",
+          "label": "Alpaca"
+        },
+        {
+          "id": "alpaca_other",
+          "label": "Alpaca Other"
+        },
+        {
+          "id": "kudge",
+          "label": "KUDGE"
+        },
+        {
+          "id": "kudge_other",
+          "label": "KUDGE Other"
+        },
+        {
+          "id": "openhermes",
+          "label": "OpenHermes"
+        },
+        {
+          "id": "openhermes_other",
+          "label": "OpenHermes Other"
+        },
+        {
+          "id": "vicuna",
+          "label": "Vicuna"
+        },
+        {
+          "id": "vicuna_other",
+          "label": "Vicuna Other"
+        }
+      ]
+    },
+    {
+      "id": "ASR",
+      "label": "ASR",
+      "metricLabel": "CER (%)",
+      "shortMetric": "CER",
+      "lowerBetter": true,
+      "datasets": [
+        {
+          "id": "common_voice_korea",
+          "label": "CommonVoice-KO"
+        },
+        {
+          "id": "common_voice_korea_other",
+          "label": "CommonVoice-KO Other"
+        },
+        {
+          "id": "ksponspeech_eval_clean",
+          "label": "KsponSpeech Clean"
+        },
+        {
+          "id": "ksponspeech_eval_other",
+          "label": "KsponSpeech Other"
+        },
+        {
+          "id": "zeroth_korean_test",
+          "label": "Zeroth-Korean"
+        },
+        {
+          "id": "zeroth_korean_test_other",
+          "label": "Zeroth-Korean Other"
+        }
+      ]
+    },
+    {
+      "id": "Translation",
+      "label": "Translation",
+      "metricLabel": "BLEU / METEOR",
+      "shortMetric": "BLEU / METEOR",
+      "lowerBetter": false,
+      "datasets": [
+        {
+          "id": "etri_tst-COMMON",
+          "label": "ETRI-TST-Common"
+        },
+        {
+          "id": "etri_tst-HE",
+          "label": "ETRI-TST-HE"
+        }
+      ]
+    },
+    {
+      "id": "LSQA",
+      "label": "Long Speech Understanding",
+      "metricLabel": "Accuracy (%)",
+      "shortMetric": "Acc(%)",
+      "lowerBetter": false,
+      "datasets": [
+        {
+          "id": "mctest",
+          "label": "MCTest"
+        },
+        {
+          "id": "mctest_other",
+          "label": "MCTest Other"
+        }
+      ]
+    }
+  ],
+  "entries": [
+    {
+      "id": "gemini_flash",
+      "rank_name": "gemini_flash",
+      "model": "gemini-2.5-flash-lite",
+      "url": "",
+      "tasks": {
+        "K-disentQA": {
+          "history_after_chosun": {
+            "value": 64.63414634146342,
+            "display": "64.63"
+          },
+          "history_after_chosun_other": {
+            "value": 58.536585365853654,
+            "display": "58.54"
+          },
+          "history_before_chosun": {
+            "value": 66.33663366336634,
+            "display": "66.34"
+          },
+          "history_before_chosun_other": {
+            "value": 65.34653465346535,
+            "display": "65.35"
+          },
+          "k-sports": {
+            "value": 86.36363636363636,
+            "display": "86.36"
+          },
+          "k-sports_other": {
+            "value": 84.0909090909091,
+            "display": "84.09"
+          },
+          "kpop": {
+            "value": 69.90291262135922,
+            "display": "69.90"
+          },
+          "kpop_other": {
+            "value": 73.7864077669903,
+            "display": "73.79"
+          }
+        },
+        "SQA": {
+          "click": {
+            "value": 60.48053024026512,
+            "display": "60.48"
+          },
+          "click_other": {
+            "value": 60.56338028169014,
+            "display": "60.56"
+          },
+          "kobest_boolq": {
+            "value": 51.967799642218246,
+            "display": "51.97"
+          },
+          "kobest_boolq_other": {
+            "value": 50.44722719141323,
+            "display": "50.45"
+          }
+        },
+        "Instruct": {
+          "alpaca": {
+            "value": 85.94202898550726,
+            "display": "85.94"
+          },
+          "alpaca_other": {
+            "value": 86.8840579710145,
+            "display": "86.88"
+          },
+          "kudge": {
+            "value": 70.28725314183116,
+            "display": "70.29"
+          },
+          "kudge_other": {
+            "value": 70.3859964093356,
+            "display": "70.39"
+          },
+          "openhermes": {
+            "value": 82.69230769230771,
+            "display": "82.69"
+          },
+          "openhermes_other": {
+            "value": 80.70512820512825,
+            "display": "80.71"
+          },
+          "vicuna": {
+            "value": 76.42857142857144,
+            "display": "76.43"
+          },
+          "vicuna_other": {
+            "value": 73.28571428571429,
+            "display": "73.29"
+          }
+        },
+        "ASR": {
+          "common_voice_korea": {
+            "value": 13.738049713193117,
+            "display": "13.74"
+          },
+          "common_voice_korea_other": {
+            "value": 26.73996175908222,
+            "display": "26.74"
+          },
+          "ksponspeech_eval_clean": {
+            "value": 83.18526725056962,
+            "display": "83.19"
+          },
+          "ksponspeech_eval_other": {
+            "value": 45.13922315086276,
+            "display": "45.14"
+          },
+          "zeroth_korean_test": {
+            "value": 13.599128992119452,
+            "display": "13.60"
+          },
+          "zeroth_korean_test_other": {
+            "value": 14.558274574865202,
+            "display": "14.56"
+          }
+        },
+        "Translation": {
+          "etri_tst-COMMON": {
+            "value": 22.879423047941344,
+            "display": "22.88 / 47.15"
+          },
+          "etri_tst-HE": {
+            "value": 26.953160916836307,
+            "display": "26.95 / 52.59"
+          }
+        },
+        "LSQA": {
+          "mctest": {
+            "value": 93.88379204892966,
+            "display": "93.88"
+          },
+          "mctest_other": {
+            "value": 94.18960244648318,
+            "display": "94.19"
+          }
+        }
+      }
+    },
+    {
+      "id": "gemma3n_vllm",
+      "rank_name": "gemma3n_vllm",
+      "model": "google/gemma-3n-E4B-it",
+      "url": "",
+      "tasks": {
+        "K-disentQA": {
+          "history_after_chosun": {
+            "value": 50.0,
+            "display": "50.00"
+          },
+          "history_after_chosun_other": {
+            "value": 48.78048780487805,
+            "display": "48.78"
+          },
+          "history_before_chosun": {
+            "value": 69.3069306930693,
+            "display": "69.31"
+          },
+          "history_before_chosun_other": {
+            "value": 68.31683168316832,
+            "display": "68.32"
+          },
+          "k-sports": {
+            "value": 76.13636363636364,
+            "display": "76.14"
+          },
+          "k-sports_other": {
+            "value": 73.86363636363636,
+            "display": "73.86"
+          },
+          "kpop": {
+            "value": 65.0485436893204,
+            "display": "65.05"
+          },
+          "kpop_other": {
+            "value": 64.07766990291263,
+            "display": "64.08"
+          }
+        },
+        "SQA": {
+          "click": {
+            "value": 35.294117647058826,
+            "display": "35.29"
+          },
+          "click_other": {
+            "value": 35.708367854183926,
+            "display": "35.71"
+          },
+          "kobest_boolq": {
+            "value": 50.715563506261184,
+            "display": "50.72"
+          },
+          "kobest_boolq_other": {
+            "value": 51.25223613595706,
+            "display": "51.25"
+          }
+        },
+        "Instruct": {
+          "alpaca": {
+            "value": 82.97101449275362,
+            "display": "82.97"
+          },
+          "alpaca_other": {
+            "value": 83.36231884057973,
+            "display": "83.36"
+          },
+          "kudge": {
+            "value": 71.38240574506275,
+            "display": "71.38"
+          },
+          "kudge_other": {
+            "value": 70.69120287253139,
+            "display": "70.69"
+          },
+          "openhermes": {
+            "value": 84.61538461538464,
+            "display": "84.62"
+          },
+          "openhermes_other": {
+            "value": 85.96153846153848,
+            "display": "85.96"
+          },
+          "vicuna": {
+            "value": 80.21428571428574,
+            "display": "80.21"
+          },
+          "vicuna_other": {
+            "value": 80.00000000000003,
+            "display": "80.00"
+          }
+        },
+        "ASR": {
+          "common_voice_korea": {
+            "value": 144.5793499043977,
+            "display": "144.58"
+          },
+          "common_voice_korea_other": {
+            "value": 178.34163155878971,
+            "display": "178.34"
+          },
+          "ksponspeech_eval_clean": {
+            "value": 142.99420974518696,
+            "display": "142.99"
+          },
+          "ksponspeech_eval_other": {
+            "value": 130.76182558088652,
+            "display": "130.76"
+          },
+          "zeroth_korean_test": {
+            "value": 107.88054749066778,
+            "display": "107.88"
+          },
+          "zeroth_korean_test_other": {
+            "value": 118.62297801742017,
+            "display": "118.62"
+          }
+        },
+        "Translation": {
+          "etri_tst-COMMON": {
+            "value": 5.1704245727847375,
+            "display": "5.17 / 16.97"
+          },
+          "etri_tst-HE": {
+            "value": 5.235146756138342,
+            "display": "5.24 / 17.50"
+          }
+        },
+        "LSQA": {
+          "mctest": {
+            "value": 47.706422018348626,
+            "display": "47.71"
+          },
+          "mctest_other": {
+            "value": 48.62385321100918,
+            "display": "48.62"
+          }
+        }
+      }
+    },
+    {
+      "id": "gpt_realtime_mini",
+      "rank_name": "gpt_realtime_mini",
+      "model": "gpt-audio-mini",
+      "url": "",
+      "tasks": {
+        "K-disentQA": {
+          "history_after_chosun": {
+            "value": 45.1219512195122,
+            "display": "45.12"
+          },
+          "history_after_chosun_other": {
+            "value": 42.68292682926829,
+            "display": "42.68"
+          },
+          "history_before_chosun": {
+            "value": 64.35643564356435,
+            "display": "64.36"
+          },
+          "history_before_chosun_other": {
+            "value": 62.37623762376238,
+            "display": "62.38"
+          },
+          "k-sports": {
+            "value": 45.45454545454545,
+            "display": "45.45"
+          },
+          "k-sports_other": {
+            "value": 46.590909090909086,
+            "display": "46.59"
+          },
+          "kpop": {
+            "value": 41.74757281553398,
+            "display": "41.75"
+          },
+          "kpop_other": {
+            "value": 40.77669902912621,
+            "display": "40.78"
+          }
+        },
+        "SQA": {
+          "click": {
+            "value": 61.64043082021541,
+            "display": "61.64"
+          },
+          "click_other": {
+            "value": 60.06628003314002,
+            "display": "60.07"
+          },
+          "kobest_boolq": {
+            "value": 51.878354203935594,
+            "display": "51.88"
+          },
+          "kobest_boolq_other": {
+            "value": 50.44722719141323,
+            "display": "50.45"
+          }
+        },
+        "Instruct": {
+          "alpaca": {
+            "value": 90.57971014492755,
+            "display": "90.58"
+          },
+          "alpaca_other": {
+            "value": 90.57971014492755,
+            "display": "90.58"
+          },
+          "kudge": {
+            "value": 74.06642728904846,
+            "display": "74.07"
+          },
+          "kudge_other": {
+            "value": 73.98563734290842,
+            "display": "73.99"
+          },
+          "openhermes": {
+            "value": 89.42307692307693,
+            "display": "89.42"
+          },
+          "openhermes_other": {
+            "value": 89.61538461538464,
+            "display": "89.62"
+          },
+          "vicuna": {
+            "value": 82.14285714285717,
+            "display": "82.14"
+          },
+          "vicuna_other": {
+            "value": 81.78571428571429,
+            "display": "81.79"
+          }
+        },
+        "ASR": {
+          "common_voice_korea": {
+            "value": 33.04971319311664,
+            "display": "33.05"
+          },
+          "common_voice_korea_other": {
+            "value": 36.21414913957935,
+            "display": "36.21"
+          },
+          "ksponspeech_eval_clean": {
+            "value": 134.18967787788205,
+            "display": "134.19"
+          },
+          "ksponspeech_eval_other": {
+            "value": 63.6444522236322,
+            "display": "63.64"
+          },
+          "zeroth_korean_test": {
+            "value": 100.0,
+            "display": "100.00"
+          },
+          "zeroth_korean_test_other": {
+            "value": 100.0,
+            "display": "100.00"
+          }
+        },
+        "Translation": {
+          "etri_tst-COMMON": {
+            "value": 26.06340955434567,
+            "display": "26.06 / 49.37"
+          },
+          "etri_tst-HE": {
+            "value": 28.984282512215568,
+            "display": "28.98 / 52.96"
+          }
+        },
+        "LSQA": {
+          "mctest": {
+            "value": 84.09785932721712,
+            "display": "84.10"
+          },
+          "mctest_other": {
+            "value": 82.87461773700305,
+            "display": "82.87"
+          }
+        }
+      }
+    },
+    {
+      "id": "qwen",
+      "rank_name": "qwen",
+      "model": "qwen",
+      "url": "",
+      "tasks": {
+        "K-disentQA": {
+          "history_after_chosun": {
+            "value": 67.07317073170732,
+            "display": "67.07"
+          },
+          "history_after_chosun_other": {
+            "value": 63.41463414634146,
+            "display": "63.41"
+          },
+          "history_before_chosun": {
+            "value": 72.27722772277228,
+            "display": "72.28"
+          },
+          "history_before_chosun_other": {
+            "value": 66.33663366336634,
+            "display": "66.34"
+          },
+          "k-sports": {
+            "value": 88.63636363636364,
+            "display": "88.64"
+          },
+          "k-sports_other": {
+            "value": 87.5,
+            "display": "87.50"
+          },
+          "kpop": {
+            "value": 81.55339805825243,
+            "display": "81.55"
+          },
+          "kpop_other": {
+            "value": 81.55339805825243,
+            "display": "81.55"
+          }
+        },
+        "SQA": {
+          "click": {
+            "value": 33.471416735708374,
+            "display": "33.47"
+          },
+          "click_other": {
+            "value": 32.64291632145816,
+            "display": "32.64"
+          },
+          "kobest_boolq": {
+            "value": 50.62611806797853,
+            "display": "50.63"
+          },
+          "kobest_boolq_other": {
+            "value": 50.62611806797853,
+            "display": "50.63"
+          }
+        },
+        "Instruct": {
+          "alpaca": {
+            "value": 55.79710144927535,
+            "display": "55.80"
+          },
+          "alpaca_other": {
+            "value": 55.43478260869564,
+            "display": "55.43"
+          },
+          "kudge": {
+            "value": 58.88888888888886,
+            "display": "58.89"
+          },
+          "kudge_other": {
+            "value": 55.25134649910226,
+            "display": "55.25"
+          },
+          "openhermes": {
+            "value": 59.679487179487175,
+            "display": "59.68"
+          },
+          "openhermes_other": {
+            "value": 59.294871794871796,
+            "display": "59.29"
+          },
+          "vicuna": {
+            "value": 63.28571428571424,
+            "display": "63.29"
+          },
+          "vicuna_other": {
+            "value": 61.49999999999998,
+            "display": "61.50"
+          }
+        },
+        "ASR": {
+          "common_voice_korea": {
+            "value": 33.55640535372849,
+            "display": "33.56"
+          },
+          "common_voice_korea_other": {
+            "value": 38.14630409804673,
+            "display": "38.15"
+          },
+          "ksponspeech_eval_clean": {
+            "value": 39.99665544848345,
+            "display": "40.00"
+          },
+          "ksponspeech_eval_other": {
+            "value": 35.66417246459254,
+            "display": "35.66"
+          },
+          "zeroth_korean_test": {
+            "value": 22.169224388220655,
+            "display": "22.17"
+          },
+          "zeroth_korean_test_other": {
+            "value": 25.036291995022815,
+            "display": "25.04"
+          }
+        },
+        "Translation": {
+          "etri_tst-COMMON": {
+            "value": 15.98267809185928,
+            "display": "15.98 / 36.78"
+          },
+          "etri_tst-HE": {
+            "value": 19.09056943080292,
+            "display": "19.09 / 40.77"
+          }
+        },
+        "LSQA": {
+          "mctest": {
+            "value": 51.07033639143731,
+            "display": "51.07"
+          },
+          "mctest_other": {
+            "value": 50.45871559633027,
+            "display": "50.46"
+          }
+        }
+      }
+    },
+    {
+      "id": "qwen3_vllm",
+      "rank_name": "qwen3_vllm",
+      "model": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
+      "url": "",
+      "tasks": {
+        "K-disentQA": {
+          "history_after_chosun": {
+            "value": 84.14634146341463,
+            "display": "84.15"
+          },
+          "history_after_chosun_other": {
+            "value": 82.92682926829268,
+            "display": "82.93"
+          },
+          "history_before_chosun": {
+            "value": 94.05940594059405,
+            "display": "94.06"
+          },
+          "history_before_chosun_other": {
+            "value": 94.05940594059405,
+            "display": "94.06"
+          },
+          "k-sports": {
+            "value": 88.63636363636364,
+            "display": "88.64"
+          },
+          "k-sports_other": {
+            "value": 89.77272727272727,
+            "display": "89.77"
+          },
+          "kpop": {
+            "value": 88.3495145631068,
+            "display": "88.35"
+          },
+          "kpop_other": {
+            "value": 85.43689320388349,
+            "display": "85.44"
+          }
+        },
+        "SQA": {
+          "click": {
+            "value": 63.96023198011599,
+            "display": "63.96"
+          },
+          "click_other": {
+            "value": 61.55758077879039,
+            "display": "61.56"
+          },
+          "kobest_boolq": {
+            "value": 51.341681574239715,
+            "display": "51.34"
+          },
+          "kobest_boolq_other": {
+            "value": 51.43112701252236,
+            "display": "51.43"
+          }
+        },
+        "Instruct": {
+          "alpaca": {
+            "value": 84.05797101449278,
+            "display": "84.06"
+          },
+          "alpaca_other": {
+            "value": 83.04347826086959,
+            "display": "83.04"
+          },
+          "kudge": {
+            "value": 71.86714542190298,
+            "display": "71.87"
+          },
+          "kudge_other": {
+            "value": 71.82226211849184,
+            "display": "71.82"
+          },
+          "openhermes": {
+            "value": 86.5384615384616,
+            "display": "86.54"
+          },
+          "openhermes_other": {
+            "value": 85.19230769230771,
+            "display": "85.19"
+          },
+          "vicuna": {
+            "value": 79.64285714285715,
+            "display": "79.64"
+          },
+          "vicuna_other": {
+            "value": 78.42857142857143,
+            "display": "78.43"
+          }
+        },
+        "ASR": {
+          "common_voice_korea": {
+            "value": 4.961759082217973,
+            "display": "4.96"
+          },
+          "common_voice_korea_other": {
+            "value": 6.779011872845652,
+            "display": "6.78"
+          },
+          "ksponspeech_eval_clean": {
+            "value": 8.459624992161208,
+            "display": "8.46"
+          },
+          "ksponspeech_eval_other": {
+            "value": 7.907058154290465,
+            "display": "7.91"
+          },
+          "zeroth_korean_test": {
+            "value": 3.3336789713811696,
+            "display": "3.33"
+          },
+          "zeroth_korean_test_other": {
+            "value": 3.9143508917461634,
+            "display": "3.91"
+          }
+        },
+        "Translation": {
+          "etri_tst-COMMON": {
+            "value": 28.528409006302443,
+            "display": "28.53 / 52.02"
+          },
+          "etri_tst-HE": {
+            "value": 31.70803814468474,
+            "display": "31.71 / 55.79"
+          }
+        },
+        "LSQA": {
+          "mctest": {
+            "value": 92.04892966360856,
+            "display": "92.05"
+          },
+          "mctest_other": {
+            "value": 95.71865443425077,
+            "display": "95.72"
+          }
+        }
+      }
+    },
+    {
+      "id": "voxtral3b",
+      "rank_name": "voxtral3b",
+      "model": "mistralai/Voxtral-Mini-3B-2507",
+      "url": "",
+      "tasks": {
+        "K-disentQA": {
+          "history_after_chosun": {
+            "value": 81.70731707317073,
+            "display": "81.71"
+          },
+          "history_after_chosun_other": {
+            "value": 78.04878048780488,
+            "display": "78.05"
+          },
+          "history_before_chosun": {
+            "value": 86.13861386138613,
+            "display": "86.14"
+          },
+          "history_before_chosun_other": {
+            "value": 86.13861386138613,
+            "display": "86.14"
+          },
+          "k-sports": {
+            "value": 94.31818181818183,
+            "display": "94.32"
+          },
+          "k-sports_other": {
+            "value": 94.31818181818183,
+            "display": "94.32"
+          },
+          "kpop": {
+            "value": 89.32038834951457,
+            "display": "89.32"
+          },
+          "kpop_other": {
+            "value": 90.29126213592234,
+            "display": "90.29"
+          }
+        },
+        "SQA": {
+          "click": {
+            "value": 41.25932062966031,
+            "display": "41.26"
+          },
+          "click_other": {
+            "value": 41.01077050538525,
+            "display": "41.01"
+          },
+          "kobest_boolq": {
+            "value": 51.788908765652955,
+            "display": "51.79"
+          },
+          "kobest_boolq_other": {
+            "value": 51.073345259391765,
+            "display": "51.07"
+          }
+        },
+        "Instruct": {
+          "alpaca": {
+            "value": 72.89855072463769,
+            "display": "72.90"
+          },
+          "alpaca_other": {
+            "value": 72.46376811594205,
+            "display": "72.46"
+          },
+          "kudge": {
+            "value": 61.9658886894074,
+            "display": "61.97"
+          },
+          "kudge_other": {
+            "value": 61.69658886894065,
+            "display": "61.70"
+          },
+          "openhermes": {
+            "value": 69.10256410256412,
+            "display": "69.10"
+          },
+          "openhermes_other": {
+            "value": 69.61538461538463,
+            "display": "69.62"
+          },
+          "vicuna": {
+            "value": 67.78571428571428,
+            "display": "67.79"
+          },
+          "vicuna_other": {
+            "value": 69.49999999999997,
+            "display": "69.50"
+          }
+        },
+        "ASR": {
+          "common_voice_korea": {
+            "value": 60.09560229445506,
+            "display": "60.10"
+          },
+          "common_voice_korea_other": {
+            "value": 58.98123324396782,
+            "display": "58.98"
+          },
+          "ksponspeech_eval_clean": {
+            "value": 62.62463680260875,
+            "display": "62.62"
+          },
+          "ksponspeech_eval_other": {
+            "value": 56.042240989512685,
+            "display": "56.04"
+          },
+          "zeroth_korean_test": {
+            "value": 40.92181667357943,
+            "display": "40.92"
+          },
+          "zeroth_korean_test_other": {
+            "value": 39.06574035669846,
+            "display": "39.07"
+          }
+        },
+        "Translation": {
+          "etri_tst-COMMON": {
+            "value": 23.06821405492274,
+            "display": "23.07 / 45.95"
+          },
+          "etri_tst-HE": {
+            "value": 26.078611466879202,
+            "display": "26.08 / 50.16"
+          }
+        },
+        "LSQA": {
+          "mctest": {
+            "value": 88.37920489296636,
+            "display": "88.38"
+          },
+          "mctest_other": {
+            "value": 89.29663608562691,
+            "display": "89.30"
+          }
+        }
+      }
+    }
+  ]
+}

data/results_real/ASR/gemini_flash/common_voice_korea/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 523,
+  "total_cer": 0.13738049713193118,
+  "total_edit_distance": 1437,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 1678.7555549144745,
+  "timestamp": "2026-03-12T02:49:36.480317"
+}

data/results_real/ASR/gemini_flash/common_voice_korea/prompt_v2/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 523,
+  "total_cer": 0.13738049713193118,
+  "total_edit_distance": 1437,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 1678.7555549144745,
+  "timestamp": "2026-03-12T02:49:36.480317"
+}

data/results_real/ASR/gemini_flash/common_voice_korea_other/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 523,
+  "total_cer": 0.2673996175908222,
+  "total_edit_distance": 2797,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 2389.414547920227,
+  "timestamp": "2026-03-12T03:29:28.746633"
+}

data/results_real/ASR/gemini_flash/common_voice_korea_other/prompt_v2/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 523,
+  "total_cer": 0.2673996175908222,
+  "total_edit_distance": 2797,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 2389.414547920227,
+  "timestamp": "2026-03-12T03:29:28.746633"
+}

data/results_real/ASR/gemini_flash/ksponspeech_eval_clean/ksponspeech_eval_clean_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_clean.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 3000,
+  "total_cer": 0.8318526725056962,
+  "total_edit_distance": 39795,
+  "total_ref_length": 47839,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 14890.70679473877,
+  "timestamp": "2026-03-12T07:37:42.309543"
+}

data/results_real/ASR/gemini_flash/ksponspeech_eval_clean/prompt_v2/ksponspeech_eval_clean_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_clean.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 3000,
+  "total_cer": 0.8318526725056962,
+  "total_edit_distance": 39795,
+  "total_ref_length": 47839,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 14890.70679473877,
+  "timestamp": "2026-03-12T07:37:42.309543"
+}

data/results_real/ASR/gemini_flash/ksponspeech_eval_other/ksponspeech_eval_other_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_other.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 3000,
+  "total_cer": 0.4513922315086276,
+  "total_edit_distance": 30947,
+  "total_ref_length": 68559,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 14233.098718166351,
+  "timestamp": "2026-03-12T11:34:58.652305"
+}

data/results_real/ASR/gemini_flash/ksponspeech_eval_other/prompt_v2/ksponspeech_eval_other_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_other.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 3000,
+  "total_cer": 0.4513922315086276,
+  "total_edit_distance": 30947,
+  "total_ref_length": 68559,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 14233.098718166351,
+  "timestamp": "2026-03-12T11:34:58.652305"
+}

data/results_real/ASR/gemini_flash/zeroth_korean_test/prompt_v2/zeroth_korean_test_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 457,
+  "total_cer": 0.13599128992119452,
+  "total_edit_distance": 2623,
+  "total_ref_length": 19288,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 2212.4609863758087,
+  "timestamp": "2026-03-12T12:11:54.141376"
+}

data/results_real/ASR/gemini_flash/zeroth_korean_test/zeroth_korean_test_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 457,
+  "total_cer": 0.13599128992119452,
+  "total_edit_distance": 2623,
+  "total_ref_length": 19288,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 2212.4609863758087,
+  "timestamp": "2026-03-12T12:11:54.141376"
+}

data/results_real/ASR/gemini_flash/zeroth_korean_test_other/prompt_v2/zeroth_korean_test_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test_noisy.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 457,
+  "total_cer": 0.14558274574865201,
+  "total_edit_distance": 2808,
+  "total_ref_length": 19288,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 2302.8534002304077,
+  "timestamp": "2026-03-12T12:50:19.843093"
+}

data/results_real/ASR/gemini_flash/zeroth_korean_test_other/zeroth_korean_test_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test_noisy.jsonl",
+  "model": "gemini-2.5-flash-lite",
+  "total_samples": 457,
+  "total_cer": 0.14558274574865201,
+  "total_edit_distance": 2808,
+  "total_ref_length": 19288,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 2302.8534002304077,
+  "timestamp": "2026-03-12T12:50:19.843093"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 523,
+  "total_cer": 1.4457934990439771,
+  "total_edit_distance": 15123,
+  "total_ref_length": 10460,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 110.93934440612793,
+  "timestamp": "2026-03-06T09:23:46.623712"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea/prompt_v1/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 523,
+  "total_cer": 2.8693116634799236,
+  "total_edit_distance": 30013,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "주어진 음성을 한국어로 전사하고 전사된 한국어 텍스트만 출력해 주세요.",
+  "elapsed_time_seconds": 307.95989894866943,
+  "timestamp": "2026-03-10T19:22:21.797930"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea/prompt_v2/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 523,
+  "total_cer": 1.6489483747609943,
+  "total_edit_distance": 17248,
+  "total_ref_length": 10460,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 134.68821096420288,
+  "timestamp": "2026-03-06T09:21:55.660171"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea/prompt_v3/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 523,
+  "total_cer": 1.4457934990439771,
+  "total_edit_distance": 15123,
+  "total_ref_length": 10460,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 110.93934440612793,
+  "timestamp": "2026-03-06T09:23:46.623712"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea/prompt_v4/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 523,
+  "total_cer": 1.8219885277246655,
+  "total_edit_distance": 19058,
+  "total_ref_length": 10460,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "맥락으로 주어진 음성을 한국어로 전사해 주시되 다른 설명 없이 전사된 한국어 텍스트만 응답해 주세요.",
+  "elapsed_time_seconds": 147.14897418022156,
+  "timestamp": "2026-03-06T09:26:13.796740"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea_other/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 522,
+  "total_cer": 1.7834163155878973,
+  "total_edit_distance": 18626,
+  "total_ref_length": 10444,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 147.34291195869446,
+  "timestamp": "2026-03-06T09:36:14.543997"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea_other/prompt_v1/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 522,
+  "total_cer": 1.945710455764075,
+  "total_edit_distance": 20321,
+  "total_ref_length": 10444,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "주어진 음성을 한국어로 전사하고 전사된 한국어 텍스트만 출력해 주세요.",
+  "elapsed_time_seconds": 185.2561948299408,
+  "timestamp": "2026-03-06T09:30:59.463344"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea_other/prompt_v2/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 522,
+  "total_cer": 2.036480275756415,
+  "total_edit_distance": 21269,
+  "total_ref_length": 10444,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 167.68726301193237,
+  "timestamp": "2026-03-06T09:33:47.176678"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea_other/prompt_v3/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 522,
+  "total_cer": 1.7834163155878973,
+  "total_edit_distance": 18626,
+  "total_ref_length": 10444,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 147.34291195869446,
+  "timestamp": "2026-03-06T09:36:14.543997"
+}

data/results_real/ASR/gemma3n_vllm/common_voice_korea_other/prompt_v4/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 522,
+  "total_cer": 2.239946380697051,
+  "total_edit_distance": 23394,
+  "total_ref_length": 10444,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "맥락으로 주어진 음성을 한국어로 전사해 주시되 다른 설명 없이 전사된 한국어 텍스트만 응답해 주세요.",
+  "elapsed_time_seconds": 191.64880776405334,
+  "timestamp": "2026-03-06T09:39:26.216829"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_clean/ksponspeech_eval_clean_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_clean.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.4299420974518697,
+  "total_edit_distance": 68407,
+  "total_ref_length": 47839,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 570.3824157714844,
+  "timestamp": "2026-03-06T10:10:14.044395"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_clean/prompt_v1/ksponspeech_eval_clean_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_clean.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.4493613996948096,
+  "total_edit_distance": 69336,
+  "total_ref_length": 47839,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "주어진 음성을 한국어로 전사하고 전사된 한국어 텍스트만 출력해 주세요.",
+  "elapsed_time_seconds": 603.2837119102478,
+  "timestamp": "2026-03-06T09:51:00.969666"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_clean/prompt_v2/ksponspeech_eval_clean_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_clean.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.5473985660235372,
+  "total_edit_distance": 74026,
+  "total_ref_length": 47839,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 582.5036385059357,
+  "timestamp": "2026-03-06T10:00:43.569889"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_clean/prompt_v3/ksponspeech_eval_clean_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_clean.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.4299420974518697,
+  "total_edit_distance": 68407,
+  "total_ref_length": 47839,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 570.3824157714844,
+  "timestamp": "2026-03-06T10:10:14.044395"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_clean/prompt_v4/ksponspeech_eval_clean_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_clean.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.804092894918372,
+  "total_edit_distance": 86306,
+  "total_ref_length": 47839,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "맥락으로 주어진 음성을 한국어로 전사해 주시되 다른 설명 없이 전사된 한국어 텍스트만 응답해 주세요.",
+  "elapsed_time_seconds": 712.8236339092255,
+  "timestamp": "2026-03-06T10:22:06.959856"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_other/ksponspeech_eval_other_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_other.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.3076182558088654,
+  "total_edit_distance": 89649,
+  "total_ref_length": 68559,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 584.8936712741852,
+  "timestamp": "2026-03-06T10:55:11.069525"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_other/prompt_v1/ksponspeech_eval_other_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_other.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.3692585947869718,
+  "total_edit_distance": 93875,
+  "total_ref_length": 68559,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "주어진 음성을 한국어로 전사하고 전사된 한국어 텍스트만 출력해 주세요.",
+  "elapsed_time_seconds": 658.2554221153259,
+  "timestamp": "2026-03-06T10:34:39.983630"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_other/prompt_v2/ksponspeech_eval_other_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_other.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.463206872912382,
+  "total_edit_distance": 100316,
+  "total_ref_length": 68559,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 646.0058970451355,
+  "timestamp": "2026-03-06T10:45:26.085344"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_other/prompt_v3/ksponspeech_eval_other_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_other.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.3076182558088654,
+  "total_edit_distance": 89649,
+  "total_ref_length": 68559,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 584.8936712741852,
+  "timestamp": "2026-03-06T10:55:11.069525"
+}

data/results_real/ASR/gemma3n_vllm/ksponspeech_eval_other/prompt_v4/ksponspeech_eval_other_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "ksponspeech_eval_other.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 3000,
+  "total_cer": 1.6952114237372191,
+  "total_edit_distance": 116222,
+  "total_ref_length": 68559,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "맥락으로 주어진 음성을 한국어로 전사해 주시되 다른 설명 없이 전사된 한국어 텍스트만 응답해 주세요.",
+  "elapsed_time_seconds": 814.8545458316803,
+  "timestamp": "2026-03-06T11:08:46.021211"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test/prompt_v1/zeroth_korean_test_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.2059829946080465,
+  "total_edit_distance": 23261,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "주어진 음성을 한국어로 전사하고 전사된 한국어 텍스트만 출력해 주세요.",
+  "elapsed_time_seconds": 121.40821027755737,
+  "timestamp": "2026-03-06T11:12:15.746632"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test/prompt_v2/zeroth_korean_test_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.150819162173372,
+  "total_edit_distance": 22197,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 85.42956852912903,
+  "timestamp": "2026-03-06T11:13:41.191670"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test/prompt_v3/zeroth_korean_test_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.0788054749066778,
+  "total_edit_distance": 20808,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 90.69133472442627,
+  "timestamp": "2026-03-06T11:15:11.905767"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test/prompt_v4/zeroth_korean_test_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.4709664039817503,
+  "total_edit_distance": 28372,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "맥락으로 주어진 음성을 한국어로 전사해 주시되 다른 설명 없이 전사된 한국어 텍스트만 응답해 주세요.",
+  "elapsed_time_seconds": 155.58400297164917,
+  "timestamp": "2026-03-06T11:17:47.512263"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test/zeroth_korean_test_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.0788054749066778,
+  "total_edit_distance": 20808,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 90.69133472442627,
+  "timestamp": "2026-03-06T11:15:11.905767"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test_other/prompt_v1/zeroth_korean_test_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.227447117378681,
+  "total_edit_distance": 23675,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "주어진 음성을 한국어로 전사하고 전사된 한국어 텍스트만 출력해 주세요.",
+  "elapsed_time_seconds": 128.4206943511963,
+  "timestamp": "2026-03-06T11:21:24.146182"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test_other/prompt_v2/zeroth_korean_test_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.294068851099129,
+  "total_edit_distance": 24960,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 118.1246964931488,
+  "timestamp": "2026-03-06T11:23:22.295773"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test_other/prompt_v3/zeroth_korean_test_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.1862297801742017,
+  "total_edit_distance": 22880,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 110.6101644039154,
+  "timestamp": "2026-03-06T11:25:12.929083"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test_other/prompt_v4/zeroth_korean_test_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.6815636665284115,
+  "total_edit_distance": 32434,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "맥락으로 주어진 음성을 한국어로 전사해 주시되 다른 설명 없이 전사된 한국어 텍스트만 응답해 주세요.",
+  "elapsed_time_seconds": 198.1166172027588,
+  "timestamp": "2026-03-06T11:28:31.068015"
+}

data/results_real/ASR/gemma3n_vllm/zeroth_korean_test_other/zeroth_korean_test_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "zeroth_korean_test_noisy.jsonl",
+  "model": "google/gemma-3n-E4B-it",
+  "total_samples": 457,
+  "total_cer": 1.1862297801742017,
+  "total_edit_distance": 22880,
+  "total_ref_length": 19288,
+  "normalization": "숫자 변형 허용 + 공백 제거",
+  "prompt": "제공된 음성의 내용을 한국어로 정확하게 전사해 주시고 전사된 문장만 출력해 주세요.",
+  "elapsed_time_seconds": 110.6101644039154,
+  "timestamp": "2026-03-06T11:25:12.929083"
+}

data/results_real/ASR/gpt_realtime_mini/common_voice_korea/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "gpt-audio-mini",
+  "total_samples": 523,
+  "total_cer": 0.33049713193116637,
+  "total_edit_distance": 3457,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 480.61955213546753,
+  "timestamp": "2026-03-11T18:28:40.586049"
+}

data/results_real/ASR/gpt_realtime_mini/common_voice_korea/prompt_v2/common_voice_korea_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea.jsonl",
+  "model": "gpt-audio-mini",
+  "total_samples": 523,
+  "total_cer": 0.33049713193116637,
+  "total_edit_distance": 3457,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 480.61955213546753,
+  "timestamp": "2026-03-11T18:28:40.586049"
+}

data/results_real/ASR/gpt_realtime_mini/common_voice_korea_other/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "gpt-audio-mini",
+  "total_samples": 523,
+  "total_cer": 0.3621414913957935,
+  "total_edit_distance": 3788,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 486.3854024410248,
+  "timestamp": "2026-03-11T18:36:47.342577"
+}

data/results_real/ASR/gpt_realtime_mini/common_voice_korea_other/prompt_v2/common_voice_korea_noisy_summary.json

@@ -0,0 +1,12 @@
+{
+  "dataset": "common_voice_korea_noisy.jsonl",
+  "model": "gpt-audio-mini",
+  "total_samples": 523,
+  "total_cer": 0.3621414913957935,
+  "total_edit_distance": 3788,
+  "total_ref_length": 10460,
+  "normalization": "구두점 제거 + 공백 제거",
+  "prompt": "다음 음성을 한국어로 전사해 주시고 설명이나 추가 문장 없이 전사 결과만 제공해 주세요.",
+  "elapsed_time_seconds": 486.3854024410248,
+  "timestamp": "2026-03-11T18:36:47.342577"
+}