Michael Rabinovich commited on
Commit ·
0bd3c0a
1
Parent(s): 7655daf
admin: promote/demote helpers gated by CADGENBENCH_ADMINS
Browse filesadds admin.py with is_admin, promote_row and demote_row. promotion sets
validation_status to validated and records the evidence type; demotion
reverses it and nulls validation_method. row writes reuse
submit._hub_rmw_results so the lock and read-modify-write semantics
match the submit path. the admin set is read from the CADGENBENCH_ADMINS
space variable.
admin.py
ADDED
|
@@ -0,0 +1,112 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Admin-tab handlers for the CADGenBench leaderboard Space.
|
| 2 |
+
|
| 3 |
+
Bundle 7: promote a row into the validated tier (recording the evidence
|
| 4 |
+
type) and demote it back. Gating is the ``CADGENBENCH_ADMINS`` Space
|
| 5 |
+
variable (comma-separated HF usernames); :func:`is_admin` is the single
|
| 6 |
+
predicate the UI uses to enable or disable the controls.
|
| 7 |
+
|
| 8 |
+
Row writes go through :func:`submit._hub_rmw_results`, so the
|
| 9 |
+
``_HUB_LOCK`` + read-modify-write semantics match the submit path
|
| 10 |
+
exactly. There is no second writer of ``results.jsonl`` with its own
|
| 11 |
+
locking story.
|
| 12 |
+
"""
|
| 13 |
+
from __future__ import annotations
|
| 14 |
+
|
| 15 |
+
import logging
|
| 16 |
+
import os
|
| 17 |
+
from typing import Any
|
| 18 |
+
|
| 19 |
+
import gradio as gr
|
| 20 |
+
|
| 21 |
+
from submit import _hub_rmw_results
|
| 22 |
+
|
| 23 |
+
logger = logging.getLogger(__name__)
|
| 24 |
+
|
| 25 |
+
ADMINS_ENV = "CADGENBENCH_ADMINS"
|
| 26 |
+
|
| 27 |
+
# The evidence types accepted on promotion. Mirrors the
|
| 28 |
+
# `validation_method` enum in cadgenbench-submissions/schema.md and the
|
| 29 |
+
# validation policy doc.
|
| 30 |
+
VALID_METHODS: tuple[str, ...] = ("code", "traces", "api", "manual")
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
def admin_usernames() -> set[str]:
|
| 34 |
+
"""Parse ``CADGENBENCH_ADMINS`` into a set of HF usernames.
|
| 35 |
+
|
| 36 |
+
Comma-separated, whitespace-trimmed, empties dropped. Read fresh on
|
| 37 |
+
each call so flipping the Space variable takes effect without a code
|
| 38 |
+
deploy. Empty or unset yields an empty set, which means no one is an
|
| 39 |
+
admin and the controls stay inert.
|
| 40 |
+
"""
|
| 41 |
+
raw = os.environ.get(ADMINS_ENV, "")
|
| 42 |
+
return {part.strip() for part in raw.split(",") if part.strip()}
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
def is_admin(profile: gr.OAuthProfile | None) -> bool:
|
| 46 |
+
"""Return whether *profile* is a logged-in user in the admin set.
|
| 47 |
+
|
| 48 |
+
Logged-out users (``profile is None``) are never admins. With an
|
| 49 |
+
empty admin set no profile qualifies, so the admin controls remain
|
| 50 |
+
disabled for everyone until ``CADGENBENCH_ADMINS`` is populated.
|
| 51 |
+
"""
|
| 52 |
+
if profile is None:
|
| 53 |
+
return False
|
| 54 |
+
return profile.username in admin_usernames()
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
def promote_row(submission_id: str, method: str) -> None:
|
| 58 |
+
"""Move a row into the validated tier with the given evidence type.
|
| 59 |
+
|
| 60 |
+
Sets ``validation_status`` to ``"validated"`` and
|
| 61 |
+
``validation_method`` to *method*. Idempotent: re-promoting an
|
| 62 |
+
already-validated row to the same method lands the same values.
|
| 63 |
+
|
| 64 |
+
Raises:
|
| 65 |
+
ValueError: *method* is not one of :data:`VALID_METHODS`.
|
| 66 |
+
LookupError: no row in ``results.jsonl`` carries *submission_id*.
|
| 67 |
+
"""
|
| 68 |
+
if method not in VALID_METHODS:
|
| 69 |
+
raise ValueError(
|
| 70 |
+
f"Unknown validation_method {method!r}; expected one of "
|
| 71 |
+
f"{', '.join(VALID_METHODS)}."
|
| 72 |
+
)
|
| 73 |
+
|
| 74 |
+
def mutate(rows: list[dict[str, Any]]) -> None:
|
| 75 |
+
for row in rows:
|
| 76 |
+
if row.get("submission_id") == submission_id:
|
| 77 |
+
row["validation_status"] = "validated"
|
| 78 |
+
row["validation_method"] = method
|
| 79 |
+
return
|
| 80 |
+
raise LookupError(
|
| 81 |
+
f"No row with submission_id={submission_id!r} in results.jsonl."
|
| 82 |
+
)
|
| 83 |
+
|
| 84 |
+
_hub_rmw_results(
|
| 85 |
+
mutate,
|
| 86 |
+
commit_message=f"promote {submission_id} to validated ({method})",
|
| 87 |
+
)
|
| 88 |
+
|
| 89 |
+
|
| 90 |
+
def demote_row(submission_id: str) -> None:
|
| 91 |
+
"""Return a row to the unvalidated tier, clearing ``validation_method``.
|
| 92 |
+
|
| 93 |
+
Sets ``validation_status`` to ``"unvalidated"`` and nulls
|
| 94 |
+
``validation_method``. Idempotent on an already-unvalidated row.
|
| 95 |
+
|
| 96 |
+
Raises:
|
| 97 |
+
LookupError: no row in ``results.jsonl`` carries *submission_id*.
|
| 98 |
+
"""
|
| 99 |
+
def mutate(rows: list[dict[str, Any]]) -> None:
|
| 100 |
+
for row in rows:
|
| 101 |
+
if row.get("submission_id") == submission_id:
|
| 102 |
+
row["validation_status"] = "unvalidated"
|
| 103 |
+
row["validation_method"] = None
|
| 104 |
+
return
|
| 105 |
+
raise LookupError(
|
| 106 |
+
f"No row with submission_id={submission_id!r} in results.jsonl."
|
| 107 |
+
)
|
| 108 |
+
|
| 109 |
+
_hub_rmw_results(
|
| 110 |
+
mutate,
|
| 111 |
+
commit_message=f"demote {submission_id} to unvalidated",
|
| 112 |
+
)
|