Fix: read from leaderboard.json instead of leaderboard_space.json

app.py

@@ -0,0 +1,126 @@
+"""Streamlit app to display the NPU Arabic leaderboard."""
+
+from __future__ import annotations
+
+import json
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import List, Sequence
+
+import streamlit as st
+
+# Use the aggregated space JSON which includes score and quality_overall
+# On HuggingFace, this is uploaded as leaderboard.json (aggregated version)
+_DATA_PATH = Path("leaderboard.json")
+
+# Column order for display - score and quality_overall are prominent
+_COLUMNS: Sequence[str] = (
+    "model_name",
+    "status",
+    "score",
+    "quality_overall",
+    "avg_tps",
+    "mlqa_ar_ar_f1",
+    "xquad_ar_f1",
+    "iwslt2017-en-ar_sacrebleu",
+    "xlsum_title_ar_rougeL",
+    "xlsum_summary_ar_rougeLsum",
+    "arabic_mmlu_acc",
+    "timestamp",
+)
+_METRIC_COLUMNS: Sequence[str] = tuple(
+    col for col in _COLUMNS if col not in {"model_name", "status", "timestamp"}
+)
+
+
+def _load_rows() -> List[dict]:
+    if not _DATA_PATH.exists():
+        return []
+    try:
+        raw = json.loads(_DATA_PATH.read_text(encoding="utf-8"))
+    except json.JSONDecodeError:
+        return []
+
+    if isinstance(raw, dict):
+        data = [raw]
+    elif isinstance(raw, list):
+        data = [item for item in raw if isinstance(item, dict)]
+    else:
+        data = []
+
+    # Filter to desired columns
+    filtered: List[dict] = []
+    for row in data:
+        compact = {key: row.get(key) for key in _COLUMNS}
+        status = compact.get("status")
+        if status is None:
+            status = "Completed"
+            compact["status"] = status
+        if status != "Completed":
+            for metric_col in _METRIC_COLUMNS:
+                compact[metric_col] = float("nan")
+        filtered.append(compact)
+
+    # Sort by score (highest first), then by timestamp for ties
+    def _sort_key(item: dict) -> tuple:
+        score = item.get("score")
+        score_val = float(score) if score is not None else -1.0
+        stamp = item.get("timestamp")
+        try:
+            parsed = datetime.fromisoformat(str(stamp))
+            if parsed.tzinfo is None:
+                parsed = parsed.replace(tzinfo=timezone.utc)
+            else:
+                parsed = parsed.astimezone(timezone.utc)
+        except Exception:
+            parsed = datetime.min.replace(tzinfo=timezone.utc)
+        return (score_val, parsed)
+
+    filtered.sort(key=_sort_key, reverse=True)
+    return filtered
+
+
+# Column display names for better readability
+_COLUMN_LABELS = {
+    "model_name": "Model",
+    "status": "Status",
+    "score": "Score",
+    "quality_overall": "Quality",
+    "avg_tps": "Tokens/sec",
+    "mlqa_ar_ar_f1": "MLQA F1",
+    "xquad_ar_f1": "XQuAD F1",
+    "iwslt2017-en-ar_sacrebleu": "IWSLT BLEU",
+    "xlsum_title_ar_rougeL": "XLSum Title",
+    "xlsum_summary_ar_rougeLsum": "XLSum Summary",
+    "arabic_mmlu_acc": "MMLU Acc",
+    "timestamp": "Last Updated",
+}
+
+st.set_page_config(page_title="Intel NPU Arabic Leaderboard", layout="wide")
+st.title("🏆 Intel® NPU Arabic Leaderboard")
+
+st.markdown("""
+**Score** = √(Quality × Speed) - balances model quality with inference speed on Intel NPU.
+
+**Quality** = Average of all benchmark scores (0-100 scale).
+""")
+
+rows = _load_rows()
+if not rows:
+    st.info("No evaluations uploaded yet. Trigger a run to populate the leaderboard.")
+else:
+    st.write(
+        "Submit your model for evaluation by emailing: **model=your-hf-model-id**"
+    )
+    st.dataframe(
+        rows,
+        column_config={
+            col: st.column_config.NumberColumn(_COLUMN_LABELS.get(col, col), format="%.2f")
+            if col in _METRIC_COLUMNS
+            else st.column_config.TextColumn(_COLUMN_LABELS.get(col, col))
+            for col in _COLUMNS
+        },
+        hide_index=True,
+    )
+
+st.caption("Data auto-synced from leaderboard.json produced by the evaluation pipeline.")