eval_job: publish report + gallery in one commit

Batch the whole-submission report + per-fixture gallery thumbnails into a
single create_commit instead of one upload_file per file, matching the
sharded merge path. Faster publish and no 429 "concurrency queue"
failures on the dataset commit endpoint.

Co-authored-by: Cursor <cursoragent@cursor.com>

eval_job.py

@@ -60,7 +60,7 @@ import zipfile
 from pathlib import Path
 from typing import Any
 
-from huggingface_hub import HfApi, hf_hub_download
+from huggingface_hub import CommitOperationAdd, HfApi, hf_hub_download
 
 
 RUN_DIR = Path("/tmp/run")
@@ -161,10 +161,9 @@ def main() -> int:
     html_path = REPORT_HTML_DIR / f"{submission_id}.html"
     _run_report(RUN_DIR, html_path)
     report_json = _build_report_json(RUN_DIR)
-    _upload_reports(
-        submission_id, html_path, report_json, submissions_repo, token,
+    _publish_reports_and_gallery(
+        submission_id, html_path, report_json, RUN_DIR, submissions_repo, token,
     )
-    _upload_gallery_renders(submission_id, RUN_DIR, submissions_repo, token)
     print(f"[eval_job] done: {submission_id}", flush=True)
     return 0
 
@@ -367,84 +366,69 @@ def _build_report_json(run_dir: Path) -> dict[str, Any]:
     return {"run_summary": summary, "per_fixture_results": per_fixture}
 
 
-def _upload_reports(
+def _publish_reports_and_gallery(
     submission_id: str,
     html_path: Path,
     report_json: dict[str, Any],
-    submissions_repo: str,
-    token: str,
-) -> None:
-    """Upload ``reports/<id>.html`` + ``reports/<id>.json`` to the Hub."""
-    api = HfApi(token=token)
-    api.upload_file(
-        path_or_fileobj=str(html_path),
-        path_in_repo=f"{REPORTS_DIR_IN_REPO}/{submission_id}.html",
-        repo_id=submissions_repo,
-        repo_type="dataset",
-        commit_message=f"add HTML report for {submission_id}",
-    )
-    api.upload_file(
-        path_or_fileobj=json.dumps(
-            report_json, ensure_ascii=False, indent=2,
-        ).encode("utf-8"),
-        path_in_repo=f"{REPORTS_DIR_IN_REPO}/{submission_id}.json",
-        repo_id=submissions_repo,
-        repo_type="dataset",
-        commit_message=f"add JSON report for {submission_id}",
-    )
-    print(
-        f"[eval_job] uploaded reports/{submission_id}.{{html,json}}",
-        flush=True,
-    )
-
-
-def _upload_gallery_renders(
-    submission_id: str,
     run_dir: Path,
     submissions_repo: str,
     token: str,
 ) -> None:
-    """Upload one ``iso`` thumbnail per fixture for the leaderboard gallery.
-
-    Stages every ``<run_dir>/<fixture>/renders/iso.png`` as
-    ``renders/<id>/<fixture>.png`` in the submissions dataset (one
-    commit). These are the standalone PNGs the gallery's
-    ``renderFor()`` points at; the full multi-view renders stay
-    base64-embedded in ``reports/<id>.html`` for the self-contained
-    report. A fixture with no ``iso.png`` (missing output, or a render
-    that never ran) is simply skipped, the gallery reads the
+    """Publish report HTML/JSON + every gallery thumbnail in one commit.
+
+    Stages ``reports/<id>.{html,json}`` plus one
+    ``renders/<id>/<fixture>.png`` per fixture that produced an ``iso``
+    thumbnail, then pushes them all in a single ``create_commit``. A
+    commit-per-file both serialises the publish and hammers the
+    dataset's commit endpoint (the 429 "concurrency queue" failures that
+    stranded earlier runs); one commit is atomic, fast, and rate-limit
+    friendly. The standalone PNGs back the gallery's ``renderFor()``; the
+    full multi-view renders stay base64-embedded in ``reports/<id>.html``
+    for the self-contained report. A fixture with no ``iso.png`` (missing
+    output, or a render that never ran) is skipped; the gallery reads the
     per-fixture status from the row and draws the dashed "invalid
     generation" cell, so a thumbnail's absence is not an error.
     """
-    staged: list[tuple[Path, str]] = []
+    operations: list[CommitOperationAdd] = [
+        CommitOperationAdd(
+            path_in_repo=f"{REPORTS_DIR_IN_REPO}/{submission_id}.html",
+            path_or_fileobj=str(html_path),
+        ),
+        CommitOperationAdd(
+            path_in_repo=f"{REPORTS_DIR_IN_REPO}/{submission_id}.json",
+            path_or_fileobj=json.dumps(
+                report_json, ensure_ascii=False, indent=2,
+            ).encode("utf-8"),
+        ),
+    ]
+    render_count = 0
     for fixture_dir in sorted(d for d in run_dir.iterdir() if d.is_dir()):
         iso_png = fixture_dir / "renders" / f"{GALLERY_THUMB_VIEW}.png"
         if iso_png.is_file():
-            staged.append((iso_png, fixture_dir.name))
-
-    if not staged:
-        print(
-            f"[eval_job] no gallery renders to upload for {submission_id}",
-            flush=True,
-        )
-        return
+            operations.append(
+                CommitOperationAdd(
+                    path_in_repo=(
+                        f"{RENDERS_DIR_IN_REPO}/{submission_id}/"
+                        f"{fixture_dir.name}.png"
+                    ),
+                    path_or_fileobj=str(iso_png),
+                )
+            )
+            render_count += 1
 
     api = HfApi(token=token)
-    for iso_png, fixture_name in staged:
-        api.upload_file(
-            path_or_fileobj=str(iso_png),
-            path_in_repo=(
-                f"{RENDERS_DIR_IN_REPO}/{submission_id}/{fixture_name}.png"
-            ),
-            repo_id=submissions_repo,
-            repo_type="dataset",
-            commit_message=(
-                f"add gallery render {fixture_name} for {submission_id}"
-            ),
-        )
+    api.create_commit(
+        repo_id=submissions_repo,
+        repo_type="dataset",
+        operations=operations,
+        commit_message=(
+            f"publish report + {render_count} gallery render(s) "
+            f"for {submission_id}"
+        ),
+    )
     print(
-        f"[eval_job] uploaded {len(staged)} gallery render(s) under "
-        f"{RENDERS_DIR_IN_REPO}/{submission_id}/",
+        f"[eval_job] published reports/{submission_id}.{{html,json}} + "
+        f"{render_count} gallery render(s) in one commit",
         flush=True,
     )