Upload app.py

app.py

@@ -1,9 +1,9 @@
-import datetime
-import html as html_lib
-import json
-import os
-import re
-from typing import Dict, List, Optional
+import datetime
+import html as html_lib
+import json
+import os
+import re
+from typing import Dict, List, Optional
 
 import gradio as gr
 from datasets import Dataset, load_dataset
@@ -15,12 +15,12 @@ from content import (
     format_log,
     format_warning,
 )
-from evaluator import SimpleEvaluator
-
-TOKEN = os.getenv("HF_TOKEN") or os.getenv("TOKEN")
-DATASET_REPO = "RUC-NLPIR/GISA"
-RESULTS_REPO = "RUC-NLPIR/GISA-leaderboard"
-META_FILE = "meta.jsonl"
+from evaluator import SimpleEvaluator
+
+TOKEN = os.getenv("HF_TOKEN") or os.getenv("TOKEN")
+DATASET_REPO = "RUC-NLPIR/GISA"
+RESULTS_REPO = "RUC-NLPIR/GISA-leaderboard"
+META_FILE = "encrypted_question.jsonl"
 ANSWER_DIR = "answer"
 CACHE_DIR = "cache/answers"
 SEED_FILE = os.path.join(os.path.dirname(__file__), "seed.json")
@@ -34,61 +34,61 @@ ALLOWED_TYPES = {"item", "set", "list", "table"}
 
 os.makedirs(CACHE_DIR, exist_ok=True)
 
-api = HfApi()
-evaluator = SimpleEvaluator()
-
-
-def _extract_username(profile, request: Optional[gr.Request]) -> Optional[str]:
-    """
-    Best-effort extraction of the HF username across Gradio versions.
-
-    On Hugging Face Spaces with hf_oauth enabled, Gradio can inject an OAuth profile object
-    (usually exposing `.username`). Some versions also provide `request.username`.
-    """
-
-    if profile is not None:
-        username = getattr(profile, "username", None)
-        if username:
-            return str(username)
-        # Some versions may pass a dict-like profile
-        if isinstance(profile, dict):
-            for key in ("username", "preferred_username", "name"):
-                val = profile.get(key)
-                if val:
-                    return str(val)
-
-    if request is None:
-        return None
-
-    username = getattr(request, "username", None)
-    if username:
-        return str(username)
-
-    headers = getattr(request, "headers", None)
-    if not headers:
-        return None
-
-    # Starlette Headers is case-insensitive; also tolerate plain dicts.
-    for key in (
-        "x-forwarded-user",
-        "x-hf-user",
-        "x-huggingface-user",
-        "x-user",
-    ):
-        try:
-            val = headers.get(key)
-        except Exception:
-            val = None
-        if val:
-            return str(val)
-
-    return None
-
-
-def _safe_float(val):
-    try:
-        if val is None:
-            return None
+api = HfApi()
+evaluator = SimpleEvaluator()
+
+
+def _extract_username(profile, request: Optional[gr.Request]) -> Optional[str]:
+    """
+    Best-effort extraction of the HF username across Gradio versions.
+
+    On Hugging Face Spaces with hf_oauth enabled, Gradio can inject an OAuth profile object
+    (usually exposing `.username`). Some versions also provide `request.username`.
+    """
+
+    if profile is not None:
+        username = getattr(profile, "username", None)
+        if username:
+            return str(username)
+        # Some versions may pass a dict-like profile
+        if isinstance(profile, dict):
+            for key in ("username", "preferred_username", "name"):
+                val = profile.get(key)
+                if val:
+                    return str(val)
+
+    if request is None:
+        return None
+
+    username = getattr(request, "username", None)
+    if username:
+        return str(username)
+
+    headers = getattr(request, "headers", None)
+    if not headers:
+        return None
+
+    # Starlette Headers is case-insensitive; also tolerate plain dicts.
+    for key in (
+        "x-forwarded-user",
+        "x-hf-user",
+        "x-huggingface-user",
+        "x-user",
+    ):
+        try:
+            val = headers.get(key)
+        except Exception:
+            val = None
+        if val:
+            return str(val)
+
+    return None
+
+
+def _safe_float(val):
+    try:
+        if val is None:
+            return None
         if isinstance(val, str) and not val.strip():
             return None
         return float(val)
@@ -187,106 +187,106 @@ def _rows_from_source(source) -> List[dict]:
     return rows
 
 
-def load_seed_rows() -> List[dict]:
-    if not os.path.exists(SEED_FILE):
-        return _load_seed_from_root_script()
-    try:
-        with open(SEED_FILE, "r", encoding="utf-8") as f:
-            data = json.load(f)
-        return data if isinstance(data, list) else []
-    except Exception:
-        return _load_seed_from_root_script()
-
-
-def _load_seed_from_root_script() -> List[dict]:
-    root_script = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "script.js"))
-    if not os.path.exists(root_script):
-        return []
-    try:
-        text = _load_text(root_script)
-        match = re.search(r"const\\s+data\\s*=\\s*(\\[.*?\\]);", text, re.S)
-        if not match:
-            return []
-        arr_text = match.group(1)
-        arr_text = re.sub(r"(\\w+)\\s*:", r'\"\\1\":', arr_text)
-        arr_text = re.sub(r",\\s*([}\\]])", r"\\1", arr_text)
-        data = json.loads(arr_text)
-        return data if isinstance(data, list) else []
-    except Exception:
-        return []
-
-
-def render_page() -> str:
-    page = _load_text(INDEX_HTML)
-    page = page.replace("__LEADERBOARD_DATA__", "")
-    page = page.replace("__SCRIPT__", "")
-    return page
-
-
-def _format_score(val: Optional[float]) -> str:
-    if val is None:
-        return "-"
-    try:
-        return f"{float(val):.2f}"
-    except Exception:
-        return "-"
-
-
-def _render_leaderboard_rows(data: List[dict]) -> str:
-    # Render a static table body so the leaderboard is not empty even if client JS
-    # runs before Gradio mounts the HTML.
-    if not data:
-        return '<tr><td colspan="14">No submissions yet.</td></tr>'
-
-    # Default sort: Overall desc, then date desc (best-effort).
-    def _date_key(s: str) -> int:
-        m = re.match(r"^(\\d{4})-(\\d{2})-(\\d{2})$", str(s or ""))
-        if not m:
-            return 0
-        return int(m.group(1)) * 10000 + int(m.group(2)) * 100 + int(m.group(3))
-
-    sorted_rows = sorted(
-        data,
-        key=lambda r: (
-            _safe_float(r.get("overall")) or 0.0,
-            _date_key(r.get("date")),
-        ),
-        reverse=True,
-    )
-
-    out = []
-    for idx, row in enumerate(sorted_rows, start=1):
-        model = html_lib.escape(str(row.get("model") or "-"))
-        org = html_lib.escape(str(row.get("org") or "-"))
-        framework = html_lib.escape(str(row.get("framework") or "-"))
-        date = html_lib.escape(str(row.get("date") or "-"))
-
-        out.append(
-            "\n".join(
-                [
-                    "<tr>",
-                    f"  <td>{idx}</td>",
-                    '  <td class="model-cell">',
-                    f'    <div class="model-name">{model}</div>',
-                    f'    <div class="model-org">{org}</div>',
-                    "  </td>",
-                    f"  <td>{framework}</td>",
-                    f"  <td>{date}</td>",
-                    f'  <td class="highlight-em">{_format_score(_safe_float(row.get("overall")))}</td>',
-                    f"  <td>{_format_score(_safe_float(row.get('item_em')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('set_em')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('set_f1')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('list_em')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('list_f1')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('list_order')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('table_em')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('table_row_f1')))}</td>",
-                    f"  <td>{_format_score(_safe_float(row.get('table_item_f1')))}</td>",
-                    "</tr>",
-                ]
-            )
-        )
-    return "\n".join(out)
+def load_seed_rows() -> List[dict]:
+    if not os.path.exists(SEED_FILE):
+        return _load_seed_from_root_script()
+    try:
+        with open(SEED_FILE, "r", encoding="utf-8") as f:
+            data = json.load(f)
+        return data if isinstance(data, list) else []
+    except Exception:
+        return _load_seed_from_root_script()
+
+
+def _load_seed_from_root_script() -> List[dict]:
+    root_script = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "script.js"))
+    if not os.path.exists(root_script):
+        return []
+    try:
+        text = _load_text(root_script)
+        match = re.search(r"const\\s+data\\s*=\\s*(\\[.*?\\]);", text, re.S)
+        if not match:
+            return []
+        arr_text = match.group(1)
+        arr_text = re.sub(r"(\\w+)\\s*:", r'\"\\1\":', arr_text)
+        arr_text = re.sub(r",\\s*([}\\]])", r"\\1", arr_text)
+        data = json.loads(arr_text)
+        return data if isinstance(data, list) else []
+    except Exception:
+        return []
+
+
+def render_page() -> str:
+    page = _load_text(INDEX_HTML)
+    page = page.replace("__LEADERBOARD_DATA__", "")
+    page = page.replace("__SCRIPT__", "")
+    return page
+
+
+def _format_score(val: Optional[float]) -> str:
+    if val is None:
+        return "-"
+    try:
+        return f"{float(val):.2f}"
+    except Exception:
+        return "-"
+
+
+def _render_leaderboard_rows(data: List[dict]) -> str:
+    # Render a static table body so the leaderboard is not empty even if client JS
+    # runs before Gradio mounts the HTML.
+    if not data:
+        return '<tr><td colspan="14">No submissions yet.</td></tr>'
+
+    # Default sort: Overall desc, then date desc (best-effort).
+    def _date_key(s: str) -> int:
+        m = re.match(r"^(\\d{4})-(\\d{2})-(\\d{2})$", str(s or ""))
+        if not m:
+            return 0
+        return int(m.group(1)) * 10000 + int(m.group(2)) * 100 + int(m.group(3))
+
+    sorted_rows = sorted(
+        data,
+        key=lambda r: (
+            _safe_float(r.get("overall")) or 0.0,
+            _date_key(r.get("date")),
+        ),
+        reverse=True,
+    )
+
+    out = []
+    for idx, row in enumerate(sorted_rows, start=1):
+        model = html_lib.escape(str(row.get("model") or "-"))
+        org = html_lib.escape(str(row.get("org") or "-"))
+        framework = html_lib.escape(str(row.get("framework") or "-"))
+        date = html_lib.escape(str(row.get("date") or "-"))
+
+        out.append(
+            "\n".join(
+                [
+                    "<tr>",
+                    f"  <td>{idx}</td>",
+                    '  <td class="model-cell">',
+                    f'    <div class="model-name">{model}</div>',
+                    f'    <div class="model-org">{org}</div>',
+                    "  </td>",
+                    f"  <td>{framework}</td>",
+                    f"  <td>{date}</td>",
+                    f'  <td class="highlight-em">{_format_score(_safe_float(row.get("overall")))}</td>',
+                    f"  <td>{_format_score(_safe_float(row.get('item_em')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('set_em')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('set_f1')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('list_em')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('list_f1')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('list_order')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('table_em')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('table_row_f1')))}</td>",
+                    f"  <td>{_format_score(_safe_float(row.get('table_item_f1')))}</td>",
+                    "</tr>",
+                ]
+            )
+        )
+    return "\n".join(out)
 
 
 def build_js(data: List[dict]) -> str:
@@ -404,24 +404,24 @@ def parse_jsonl(file_path: str) -> Dict[str, str]:
     return preds
 
 
-def add_new_eval(
-    model: str,
-    org: str,
-    framework: str,
-    url: str,
-    email: str,
-    file_obj,
-    profile: Optional[gr.OAuthProfile] = None,
-    request: gr.Request = None,
-):
-    if not TOKEN:
-        return format_error("Server misconfigured: HF_TOKEN is missing.")
-
-    username = _extract_username(profile, request)
-    if not username:
-        return format_warning("Please log in with HuggingFace to submit.")
-    if not model or not org:
-        return format_warning("Please provide model name and organization.")
+def add_new_eval(
+    model: str,
+    org: str,
+    framework: str,
+    url: str,
+    email: str,
+    file_obj,
+    profile: Optional[gr.OAuthProfile] = None,
+    request: gr.Request = None,
+):
+    if not TOKEN:
+        return format_error("Server misconfigured: HF_TOKEN is missing.")
+
+    username = _extract_username(profile, request)
+    if not username:
+        return format_warning("Please log in with HuggingFace to submit.")
+    if not model or not org:
+        return format_warning("Please provide model name and organization.")
     if file_obj is None:
         return format_warning("Please upload a JSONL file.")
 
@@ -495,50 +495,50 @@ def add_new_eval(
     return format_log("Submission received! Please refresh the leaderboard to see your score.")
 
 
-seed_results_if_needed()
-leaderboard_data = build_leaderboard_rows()
-css = _load_text(STYLES_CSS)
-page_html = render_page()
-rows_html = _render_leaderboard_rows(leaderboard_data)
-page_html = page_html.replace(
-    '<tbody id="leaderboard-body"></tbody>',
-    f'<tbody id="leaderboard-body">{rows_html}</tbody>',
-)
-js = build_js(leaderboard_data)
-
-with gr.Blocks() as demo:
-    gr.HTML(page_html)
-
-    with gr.Accordion("Submit your results", open=True):
-        gr.Markdown(SUBMISSION_TEXT)
-        with gr.Row():
-            with gr.Column():
-                model_text = gr.Textbox(label="Model / System")
-                org_text = gr.Textbox(label="Organization")
-                framework_text = gr.Textbox(label="Framework", value="ReAct")
-                url_text = gr.Textbox(label="Model URL", placeholder="Optional")
-            with gr.Column():
-                email_text = gr.Textbox(label="Contact email (public)")
-                file_input = gr.File(label="Upload JSONL")
-
-        with gr.Row():
-            login_btn = gr.LoginButton()
-            submit_btn = gr.Button("Submit")
-
-        result_md = gr.Markdown()
-
-        submit_btn.click(
-            add_new_eval,
-            inputs=[
-                model_text,
-                org_text,
-                framework_text,
-                url_text,
-                email_text,
-                file_input,
-            ],
-            outputs=result_md,
-        )
+seed_results_if_needed()
+leaderboard_data = build_leaderboard_rows()
+css = _load_text(STYLES_CSS)
+page_html = render_page()
+rows_html = _render_leaderboard_rows(leaderboard_data)
+page_html = page_html.replace(
+    '<tbody id="leaderboard-body"></tbody>',
+    f'<tbody id="leaderboard-body">{rows_html}</tbody>',
+)
+js = build_js(leaderboard_data)
+
+with gr.Blocks() as demo:
+    gr.HTML(page_html)
+
+    with gr.Accordion("Submit your results", open=True):
+        gr.Markdown(SUBMISSION_TEXT)
+        with gr.Row():
+            with gr.Column():
+                model_text = gr.Textbox(label="Model / System")
+                org_text = gr.Textbox(label="Organization")
+                framework_text = gr.Textbox(label="Framework", value="ReAct")
+                url_text = gr.Textbox(label="Model URL", placeholder="Optional")
+            with gr.Column():
+                email_text = gr.Textbox(label="Contact email (public)")
+                file_input = gr.File(label="Upload JSONL")
+
+        with gr.Row():
+            login_btn = gr.LoginButton()
+            submit_btn = gr.Button("Submit")
+
+        result_md = gr.Markdown()
+
+        submit_btn.click(
+            add_new_eval,
+            inputs=[
+                model_text,
+                org_text,
+                framework_text,
+                url_text,
+                email_text,
+                file_input,
+            ],
+            outputs=result_md,
+        )
 
 def _launch():
     demo.queue()