Update miniapp_leaderboard.py

miniapp_leaderboard.py

@@ -1,49 +1,263 @@
-import html
-import pandas as pd
+import datetime
+import io
+import json
+import os
+import re
+import uuid
+from urllib.parse import urlparse
+
 import gradio as gr
+import numpy as np
+import pandas as pd
+import requests
+from huggingface_hub import HfApi, hf_hub_download
+from huggingface_hub.errors import RepositoryNotFoundError
+
+APP_NAME = "miniapp"
+
+HF_TOKEN = os.environ.get("HF_TOKEN") or os.environ.get("TOKEN") or os.environ.get("HUGGINGFACE_TOKEN")
+LEADERBOARD_DATASET = (os.environ.get("LEADERBOARD_DATASET") or "").strip()
+MAX_ENTRIES = int(os.environ.get("MAX_ENTRIES", "500"))
+
+ENTRIES_PREFIX = "entries/"
 
-# 你已有的：LEADERBOARD_COLUMNS / NUMERIC_COLS / _load_entries_df / _apply_search 或 refresh 等
-# 这里假设 df 列名还是：
-# Model name, Avg, Easy, Mid, Hard, Games, Science, Tools, Humanities, Viz, Lifestyle, Submitted at, Submitter
+LEADERBOARD_COLUMNS = [
+    "Model name",
+    "Avg",
+    "Easy",
+    "Mid",
+    "Hard",
+    "Games",
+    "Science",
+    "Tools",
+    "Humanities",
+    "Viz",
+    "Lifestyle",
+    "Submitted at",
+    "Submitter",
+]
+
+NUMERIC_COLS = [
+    "Avg",
+    "Easy",
+    "Mid",
+    "Hard",
+    "Games",
+    "Science",
+    "Tools",
+    "Humanities",
+    "Viz",
+    "Lifestyle",
+]
 
-DISPLAY_COLS = [
+# Avg 最左 + 多级表头展示用
+DISPLAY_ORDER = [
     "Avg",
     "Model name",
-    "Easy", "Mid", "Hard",
-    "Games", "Science", "Tools", "Humanities", "Viz", "Lifestyle",
+    "Easy",
+    "Mid",
+    "Hard",
+    "Games",
+    "Science",
+    "Tools",
+    "Humanities",
+    "Viz",
+    "Lifestyle",
 ]
+SORTABLE_COLS = DISPLAY_ORDER[:]
+
+IN_SPACES = bool(
+    os.environ.get("SPACE_ID")
+    or os.environ.get("SPACE_REPO_NAME")
+    or os.environ.get("SPACE_AUTHOR_NAME")
+    or os.environ.get("system", "") == "spaces"
+)
+
+
+def _api() -> HfApi:
+    return HfApi(token=HF_TOKEN)
+
+
+def _is_valid_http_url(url: str) -> bool:
+    try:
+        parsed = urlparse(url)
+        return parsed.scheme in ("http", "https") and bool(parsed.netloc)
+    except Exception:
+        return False
+
+
+def _slug(s: str, max_len: int = 60) -> str:
+    s = (s or "").strip().lower()
+    s = re.sub(r"[^a-z0-9]+", "-", s)
+    s = re.sub(r"-{2,}", "-", s).strip("-")
+    return (s[:max_len] or "x")
+
+
+def _empty_df() -> pd.DataFrame:
+    return pd.DataFrame(columns=LEADERBOARD_COLUMNS)
+
+
+def _ensure_dataset_readable() -> tuple[bool, str]:
+    if not HF_TOKEN:
+        return False, "Space is missing HF_TOKEN (Secrets)."
+    if not LEADERBOARD_DATASET:
+        return False, "Space is missing LEADERBOARD_DATASET (Secrets)."
+    api = _api()
+    try:
+        api.repo_info(repo_id=LEADERBOARD_DATASET, repo_type="dataset")
+        return True, ""
+    except RepositoryNotFoundError:
+        return False, (
+            f"Dataset repo not found: {LEADERBOARD_DATASET}. "
+            "Create it first (as a dataset) or fix LEADERBOARD_DATASET."
+        )
+    except Exception:
+        return False, "Cannot access the dataset repo. Check token permissions."
 
-def _fmt(v):
+
+def _list_entry_files() -> list[str]:
+    ok, _ = _ensure_dataset_readable()
+    if not ok:
+        return []
+    api = _api()
+    try:
+        files = api.list_repo_files(repo_id=LEADERBOARD_DATASET, repo_type="dataset")
+    except Exception:
+        return []
+
+    entry_files = [f for f in files if f.startswith(ENTRIES_PREFIX) and f.endswith(".json")]
+    entry_files.sort(reverse=True)
+    return entry_files[:MAX_ENTRIES]
+
+
+def _load_entries_df() -> pd.DataFrame:
+    ok, _ = _ensure_dataset_readable()
+    if not ok:
+        return _empty_df()
+
+    rows: list[dict] = []
+    for filename in _list_entry_files():
+        try:
+            path = hf_hub_download(
+                repo_id=LEADERBOARD_DATASET,
+                repo_type="dataset",
+                filename=filename,
+                token=HF_TOKEN,
+            )
+            with open(path, "r", encoding="utf-8") as fp:
+                row = json.load(fp)
+                rows.append(row)
+        except Exception:
+            continue
+
+    if not rows:
+        return _empty_df()
+
+    df = pd.DataFrame(rows)
+    for c in LEADERBOARD_COLUMNS:
+        if c not in df.columns:
+            df[c] = ""
+
+    df = df[LEADERBOARD_COLUMNS]
+    for c in NUMERIC_COLS:
+        df[c] = pd.to_numeric(df[c], errors="coerce")
+
+    # 默认按提交时间倒序
+    df = df.sort_values(by=["Submitted at"], ascending=False, kind="stable")
+    return df
+
+
+def _parse_hf_created_at(created_at: str) -> datetime.datetime | None:
+    try:
+        if created_at.endswith("Z"):
+            created_at = created_at[:-1] + "+00:00"
+        return datetime.datetime.fromisoformat(created_at)
+    except Exception:
+        return None
+
+
+def _check_user_eligibility(username: str) -> tuple[bool, str]:
+    """
+    - Must be older than ~4 months (>= 120 days)
+    """
+    try:
+        r = requests.get(f"https://huggingface.co/api/users/{username}/overview", timeout=10)
+        r.raise_for_status()
+        created_at = r.json().get("createdAt")
+        if not created_at:
+            return False, "Cannot verify account creation date."
+        dt = _parse_hf_created_at(created_at)
+        if not dt:
+            return False, "Cannot parse account creation date."
+        now = datetime.datetime.now(datetime.timezone.utc)
+        if dt.tzinfo is None:
+            dt = dt.replace(tzinfo=datetime.timezone.utc)
+        age_days = (now - dt).days
+        if age_days < 120:
+            return False, "Account must be older than 4 months to submit."
+        return True, ""
+    except Exception:
+        return False, "Cannot verify Hugging Face account. Please try again later."
+
+
+def _submitted_today(username: str) -> bool:
+    df = _load_entries_df()
+    if df.empty:
+        return False
+    today = datetime.datetime.utcnow().date().isoformat()
+    user_rows = df[df["Submitter"].astype(str) == username]
+    if user_rows.empty:
+        return False
+    return any(str(v).startswith(today) for v in user_rows["Submitted at"].tolist())
+
+
+# ---------- HTML Leaderboard helpers ----------
+def _fmt_cell(v):
     if v is None or (isinstance(v, float) and pd.isna(v)):
         return ""
-    if isinstance(v, (int, float)):
-        return f"{v:.2f}"
+    if isinstance(v, (int, float, np.number)):
+        return f"{float(v):.2f}"
     return str(v)
 
-def _render_table(df: pd.DataFrame, sort_col: str, sort_dir: str) -> str:
-    # 三段：Avg(最左) / Model / Pass Rate(%): Difficulty + Domain
-    # 注意：这里“Avg.(%)”放最左，满足你要求
+
+def _apply_search_and_sort(df: pd.DataFrame, search_text: str, sort_col: str, sort_dir: str) -> pd.DataFrame:
+    # 搜索仅 Model name
+    s = (search_text or "").strip().lower()
+    if s:
+        df = df[df["Model name"].astype(str).str.lower().str.contains(s, na=False)]
+
+    # 排序
+    sort_col = sort_col if sort_col in df.columns else "Avg"
+    asc = sort_dir == "asc"
+    df = df.sort_values(by=[sort_col], ascending=asc, kind="stable", na_position="last")
+    return df
+
+
+def _render_leaderboard_html(df: pd.DataFrame, sort_col: str, sort_dir: str) -> str:
+    import html as _html
+
     def th(label, col=None, cls=""):
-        # 可点击排序的列
         if col:
             arrow = ""
             if col == sort_col:
                 arrow = " ▲" if sort_dir == "asc" else " ▼"
-            return f'<th class="{cls} clickable" data-col="{html.escape(col)}">{html.escape(label)}{arrow}</th>'
-        return f'<th class="{cls}">{html.escape(label)}</th>'
+            return f'<th class="clickable {cls}" data-col="{_html.escape(col)}">{_html.escape(label)}{arrow}</th>'
+        return f'<th class="{cls}">{_html.escape(label)}</th>'
 
-    rows_html = []
+    trs = []
     for _, r in df.iterrows():
         tds = []
-        for c in DISPLAY_COLS:
-            val = _fmt(r.get(c, ""))
-            tds.append(f'<td class="num" data-col="{html.escape(c)}">{html.escape(val)}</td>' if c != "Model name"
-                       else f'<td class="model" data-col="{html.escape(c)}">{html.escape(val)}</td>')
-        rows_html.append("<tr>" + "".join(tds) + "</tr>")
+        for c in DISPLAY_ORDER:
+            val = _fmt_cell(r.get(c, ""))
+            if c == "Model name":
+                tds.append(f'<td class="model">{_html.escape(val)}</td>')
+            else:
+                tds.append(f'<td class="num">{_html.escape(val)}</td>')
+        trs.append("<tr>" + "".join(tds) + "</tr>")
 
-    table = f"""
+    return f"""
     <div class="lb-wrap">
-      <table class="lb">
+      <table class="lb" id="lb_table">
         <thead>
           <tr class="top">
             {th("Avg. (%)", "Avg", "avg")}
@@ -71,130 +285,232 @@ def _render_table(df: pd.DataFrame, sort_col: str, sort_dir: str) -> str:
           </tr>
         </thead>
         <tbody>
-          {"".join(rows_html)}
+          {''.join(trs)}
         </tbody>
       </table>
     </div>
     """
-    return table
 
-def render_leaderboard(search_text: str, sort_col: str, sort_dir: str):
-    df = _load_entries_df()
-    # 搜索仅 Model name
-    s = (search_text or "").strip().lower()
-    if s:
-        df = df[df["Model name"].astype(str).str.lower().str.contains(s, na=False)]
 
-    # 排序
-    if sort_col in df.columns:
-        asc = (sort_dir == "asc")
-        df = df.sort_values(by=[sort_col], ascending=asc, kind="stable", na_position="last")
+def render_lb(search_text: str, sort_col: str, sort_dir: str) -> str:
+    df = _load_entries_df()
+    df = _apply_search_and_sort(df, search_text, sort_col, sort_dir)
+    return _render_leaderboard_html(df, sort_col, sort_dir)
 
-    return _render_table(df, sort_col, sort_dir)
 
 def toggle_sort(clicked_col: str, current_col: str, current_dir: str):
-    # 点击同一列：切换 asc/desc；点击新列：默认 desc
     clicked_col = (clicked_col or "").strip()
-    if not clicked_col:
-        return gr.update(), current_col, current_dir
-
+    if clicked_col not in SORTABLE_COLS:
+        return current_col, current_dir
     if clicked_col == current_col:
-        new_dir = "asc" if current_dir == "desc" else "desc"
-        return clicked_col, clicked_col, new_dir
-    else:
-        return clicked_col, clicked_col, "desc"
+        return current_col, ("asc" if current_dir == "desc" else "desc")
+    return clicked_col, "desc"
+
+
+# ---------- Submit ----------
+def submit(
+    model_api: str,
+    api_key: str,
+    search_text: str,
+    sort_col: str,
+    sort_dir: str,
+    profile: gr.OAuthProfile | None,
+):
+    # Login required to submit (in Spaces)
+    if IN_SPACES and (profile is None or not getattr(profile, "username", None)):
+        return "You must log in to submit.", render_lb(search_text, sort_col, sort_dir)
+
+    submitter = (getattr(profile, "username", None) if profile is not None else "local") or "anonymous"
+
+    model_api = (model_api or "").strip()
+    api_key = (api_key or "").strip()
+
+    if not model_api:
+        return "Model API URL is required.", render_lb(search_text, sort_col, sort_dir)
+    if not _is_valid_http_url(model_api):
+        return "Model API must be a valid http(s) URL.", render_lb(search_text, sort_col, sort_dir)
+    if not api_key:
+        return "API key is required.", render_lb(search_text, sort_col, sort_dir)
+
+    ok, msg = _ensure_dataset_readable()
+    if not ok:
+        return msg, render_lb(search_text, sort_col, sort_dir)
+
+    if IN_SPACES:
+        ok, msg = _check_user_eligibility(submitter)
+        if not ok:
+            return msg, render_lb(search_text, sort_col, sort_dir)
+
+    if _submitted_today(submitter):
+        return "You have already submitted today. Please try again tomorrow.", render_lb(search_text, sort_col, sort_dir)
 
+    # api_key collected but NOT stored
+    now = datetime.datetime.utcnow().replace(microsecond=0).isoformat() + "Z"
+    nonce = uuid.uuid4().hex[:8]
+    safe_user = _slug(submitter)
 
-CSS = """
-/* 搜索框：不明显，一行 */
+    host = urlparse(model_api).netloc or "unknown"
+    model_name = host
+
+    safe_model = _slug(model_name)
+    path_in_repo = f"{ENTRIES_PREFIX}{now[:10]}/{now}-{safe_user}-{safe_model}-{nonce}.json"
+
+    payload = {
+        "Model name": model_name,
+        "Avg": None,
+        "Easy": None,
+        "Mid": None,
+        "Hard": None,
+        "Games": None,
+        "Science": None,
+        "Tools": None,
+        "Humanities": None,
+        "Viz": None,
+        "Lifestyle": None,
+        "Submitted at": now,
+        "Submitter": submitter,
+        "Model API": model_api,
+    }
+
+    api = _api()
+    data = (json.dumps(payload, ensure_ascii=False, indent=2) + "\n").encode("utf-8")
+    api.upload_file(
+        repo_id=LEADERBOARD_DATASET,
+        repo_type="dataset",
+        path_or_fileobj=io.BytesIO(data),
+        path_in_repo=path_in_repo,
+        commit_message=f"miniapp: submit {submitter}/{model_name}",
+        token=HF_TOKEN,
+    )
+
+    return "Submitted successfully.", render_lb(search_text, sort_col, sort_dir)
+
+
+CSS = r"""
+/* 顶部一行：搜索不明显 */
 #topbar { display:flex; align-items:center; gap:10px; }
-#search input { border:1px solid rgba(0,0,0,.12) !important; background:rgba(0,0,0,.02) !important; }
-#search label { display:none !important; }
-#search input::placeholder { color: rgba(0,0,0,.35); }
+#searchbox { max-width: 320px; }
+#searchbox label { display:none !important; }
+#searchbox textarea, #searchbox input {
+  border: 1px solid rgba(0,0,0,.10) !important;
+  background: rgba(0,0,0,.02) !important;
+  box-shadow: none !important;
+}
+#searchbox textarea::placeholder, #searchbox input::placeholder { color: rgba(0,0,0,.35); }
 
-/* 表格整体风格接近截图 */
+/* 表格风格接近截图 */
 .lb-wrap { width: 100%; overflow-x: auto; }
 table.lb { width: 100%; border-collapse: collapse; font-family: serif; }
-table.lb thead th { text-align:center; font-weight:700; padding:10px 8px; border-bottom:1px solid #222; }
-table.lb thead tr.top th { border-top:2px solid #222; }
-table.lb thead tr.mid th { border-bottom:1px solid #777; font-weight:700; padding-top:6px; padding-bottom:6px;}
-table.lb thead tr.bot th { border-bottom:2px solid #222; padding-top:8px; padding-bottom:8px; }
+table.lb thead th { text-align:center; font-weight:700; padding:10px 8px; border-bottom:1px solid #222; white-space:nowrap; }
+table.lb thead tr.top th { border-top:2px solid #222; font-size: 22px; }
+table.lb thead tr.mid th { border-bottom:1px solid #777; font-weight:700; padding-top:6px; padding-bottom:6px; font-size: 20px; }
+table.lb thead tr.bot th { border-bottom:2px solid #222; padding-top:8px; padding-bottom:8px; font-size: 20px; }
 
-table.lb tbody td { padding:10px 10px; border-bottom:1px solid rgba(0,0,0,.08); }
-table.lb tbody td.num { text-align:right; white-space:nowrap; }
+table.lb tbody td { padding:10px 10px; border-bottom:1px solid rgba(0,0,0,.08); font-family: sans-serif; font-size: 14px; }
+table.lb tbody td.num { text-align:right; }
 table.lb tbody td.model { text-align:left; min-width:260px; }
 
-th.avg { min-width:90px; }
+th.group { font-weight:800; letter-spacing:.2px; }
+th.avg { min-width:100px; }
 th.model { text-align:left; min-width:260px; }
-th.group { font-weight:800; letter-spacing:.3px; }
 
+/* 点击排序 */
 th.clickable { user-select:none; }
 th.clickable:hover { background: rgba(0,0,0,.04); cursor:pointer; }
+
+/* 提交区紧凑 */
+#submitbox { max-width: 720px; }
 """
 
+
 with gr.Blocks(title=f"{APP_NAME} leaderboard") as demo:
-    # 顶部一行：不明显搜索 + refresh
+    gr.Markdown(f"## {APP_NAME} leaderboard")
+
     with gr.Row(elem_id="topbar"):
         search_text = gr.Textbox(
-            elem_id="search",
-            placeholder="Search model...",
+            elem_id="searchbox",
+            placeholder="Search model…",
             show_label=False,
             container=False,
             scale=1,
         )
         refresh_btn = gr.Button("Refresh", scale=0)
 
-    # 隐藏状态：当前排序列/方向
+    # 排序状态
     sort_col = gr.State("Avg")
     sort_dir = gr.State("desc")
 
-    lb_html = gr.HTML(value=render_leaderboard("", "Avg", "desc"))
-
-    # 用一个隐藏 textbox 接收“点击的列名”（由 JS 写入）
-    clicked = gr.Textbox(visible=False)
-
-    # JS：给表头绑定点击事件，把 data-col 写入隐藏组件并触发 change
-    demo.load(
-        None,
-        js="""
-        () => {
-          const root = document;
-          function bind() {
-            const ths = root.querySelectorAll('th.clickable');
-            ths.forEach(th => {
-              th.onclick = () => {
-                const col = th.getAttribute('data-col');
-                const hidden = root.querySelector('textarea, input')?.closest('.gradio-container') 
-                return col;
-              };
-            });
-          }
-          setTimeout(bind, 300);
-          return null;
-        }
-        """,
+    # HTML leaderboard
+    lb_html = gr.HTML(value=render_lb("", "Avg", "desc"))
+
+    # 用 elem_id 精确定位的隐藏输入：JS 写入点击列名
+    clicked_col = gr.Textbox(visible=False, elem_id="clicked_col")
+
+    # JS：绑定表头点击，把 data-col 写入隐藏 textbox
+    gr.HTML(
+        """
+<script>
+(function(){
+  function bindClicks(){
+    const table = document.getElementById("lb_table");
+    const hidden = document.getElementById("clicked_col");
+    if(!table || !hidden) return;
+
+    table.querySelectorAll("th.clickable").forEach(th=>{
+      th.onclick = () => {
+        const col = th.getAttribute("data-col") || "";
+        hidden.value = col;
+        hidden.dispatchEvent(new Event("input", {bubbles:true}));
+        hidden.dispatchEvent(new Event("change", {bubbles:true}));
+      };
+    });
+  }
+
+  // 首次 + 每次页面更新后都尝试重新绑定
+  const obs = new MutationObserver(()=>bindClicks());
+  obs.observe(document.body, {subtree:true, childList:true});
+  setTimeout(bindClicks, 300);
+})();
+</script>
+"""
     )
 
-    # 更稳妥的方式：用自定义 JS 直接找到 hidden textbox 并写值触发 input 事件
-    gr.HTML("""
-    <script>
-      function bindLeaderboardClicks(){
-        const host = document.querySelector('.gradio-container');
-        if(!host) return;
-        const hidden = host.querySelector('input[type="text"][data-testid="textbox"]');
-        // 上面这行在不同版本可能不稳定；因此我们下面用更可靠的：找 visible=false 的 textbox 的 input
-      }
-    </script>
-    """)
-
-    # 简化：不用上面那段不稳定的定位，改用 gradio 的内置“前端事件传参”：
-    # ——Gradio 目前没有官方 th click -> python 的直接通道，所以建议你用下面“纯后端”方案：
-    #    1) 先不做 th 点击；改用一行内很轻的 Dropdown 做排序列（你说要点列名，所以这里先给你 HTML 版本）
-    #
-    # 如果你愿意接受“点列名不做，仍然一行内轻量排序”，我可以给你完全无 JS 的版本。
-
-    # 先把搜索/刷新做好（无 JS）
-    search_text.change(render_leaderboard, inputs=[search_text, sort_col, sort_dir], outputs=[lb_html])
-    refresh_btn.click(render_leaderboard, inputs=[search_text, sort_col, sort_dir], outputs=[lb_html])
+    # 搜索/刷新：重新渲染 HTML
+    search_text.change(render_lb, inputs=[search_text, sort_col, sort_dir], outputs=[lb_html])
+    refresh_btn.click(render_lb, inputs=[search_text, sort_col, sort_dir], outputs=[lb_html])
+
+    # 点击列名：更新排序状态 + 重新渲染
+    def _on_click(col, cur_col, cur_dir, s):
+        new_col, new_dir = toggle_sort(col, cur_col, cur_dir)
+        return new_col, new_dir, render_lb(s, new_col, new_dir)
+
+    clicked_col.change(
+        _on_click,
+        inputs=[clicked_col, sort_col, sort_dir, search_text],
+        outputs=[sort_col, sort_dir, lb_html],
+    )
+
+    gr.Markdown(
+        "### Submission\n"
+        "- Submit **Model API URL** and **API key** only\n"
+        "- Requires login (in Spaces)\n"
+        "- One submission per user per day\n"
+        "- HF account must be older than 4 months\n"
+        "- API key **will not be stored**"
+    )
+
+    with gr.Column(elem_id="submitbox"):
+        model_api = gr.Textbox(label="Model API URL (required)", placeholder="https://...")
+        api_key = gr.Textbox(label="API key (required)", type="password", placeholder="Will not be stored")
+        with gr.Row():
+            gr.LoginButton()
+            submit_btn = gr.Button("Submit", variant="primary")
+        status = gr.Markdown()
+
+    submit_btn.click(
+        submit,
+        inputs=[model_api, api_key, search_text, sort_col, sort_dir],
+        outputs=[status, lb_html],
+    )
 
 demo.launch(css=CSS, ssr_mode=False)