CADGenBench / admin.py
Michael Rabinovich
admin: promote/demote helpers gated by CADGENBENCH_ADMINS
0bd3c0a
Raw
History Blame
3.91 kB
"""Admin-tab handlers for the CADGenBench leaderboard Space.
Bundle 7: promote a row into the validated tier (recording the evidence
type) and demote it back. Gating is the ``CADGENBENCH_ADMINS`` Space
variable (comma-separated HF usernames); :func:`is_admin` is the single
predicate the UI uses to enable or disable the controls.
Row writes go through :func:`submit._hub_rmw_results`, so the
``_HUB_LOCK`` + read-modify-write semantics match the submit path
exactly. There is no second writer of ``results.jsonl`` with its own
locking story.
"""
from __future__ import annotations
import logging
import os
from typing import Any
import gradio as gr
from submit import _hub_rmw_results
logger = logging.getLogger(__name__)
ADMINS_ENV = "CADGENBENCH_ADMINS"
# The evidence types accepted on promotion. Mirrors the
# `validation_method` enum in cadgenbench-submissions/schema.md and the
# validation policy doc.
VALID_METHODS: tuple[str, ...] = ("code", "traces", "api", "manual")
def admin_usernames() -> set[str]:
"""Parse ``CADGENBENCH_ADMINS`` into a set of HF usernames.
Comma-separated, whitespace-trimmed, empties dropped. Read fresh on
each call so flipping the Space variable takes effect without a code
deploy. Empty or unset yields an empty set, which means no one is an
admin and the controls stay inert.
"""
raw = os.environ.get(ADMINS_ENV, "")
return {part.strip() for part in raw.split(",") if part.strip()}
def is_admin(profile: gr.OAuthProfile | None) -> bool:
"""Return whether *profile* is a logged-in user in the admin set.
Logged-out users (``profile is None``) are never admins. With an
empty admin set no profile qualifies, so the admin controls remain
disabled for everyone until ``CADGENBENCH_ADMINS`` is populated.
"""
if profile is None:
return False
return profile.username in admin_usernames()
def promote_row(submission_id: str, method: str) -> None:
"""Move a row into the validated tier with the given evidence type.
Sets ``validation_status`` to ``"validated"`` and
``validation_method`` to *method*. Idempotent: re-promoting an
already-validated row to the same method lands the same values.
Raises:
ValueError: *method* is not one of :data:`VALID_METHODS`.
LookupError: no row in ``results.jsonl`` carries *submission_id*.
"""
if method not in VALID_METHODS:
raise ValueError(
f"Unknown validation_method {method!r}; expected one of "
f"{', '.join(VALID_METHODS)}."
)
def mutate(rows: list[dict[str, Any]]) -> None:
for row in rows:
if row.get("submission_id") == submission_id:
row["validation_status"] = "validated"
row["validation_method"] = method
return
raise LookupError(
f"No row with submission_id={submission_id!r} in results.jsonl."
)
_hub_rmw_results(
mutate,
commit_message=f"promote {submission_id} to validated ({method})",
)
def demote_row(submission_id: str) -> None:
"""Return a row to the unvalidated tier, clearing ``validation_method``.
Sets ``validation_status`` to ``"unvalidated"`` and nulls
``validation_method``. Idempotent on an already-unvalidated row.
Raises:
LookupError: no row in ``results.jsonl`` carries *submission_id*.
"""
def mutate(rows: list[dict[str, Any]]) -> None:
for row in rows:
if row.get("submission_id") == submission_id:
row["validation_status"] = "unvalidated"
row["validation_method"] = None
return
raise LookupError(
f"No row with submission_id={submission_id!r} in results.jsonl."
)
_hub_rmw_results(
mutate,
commit_message=f"demote {submission_id} to unvalidated",
)