Label latest WVS family releases from saved catalog

Co-Authored-By: PI[openai-codex] <288921227+claudypoo@users.noreply.github.com>
This commit is contained in:
wassname
2026-09-17 21:26:12 +08:00
co-authored by PI[openai-codex]
parent a18f0ec457
commit f2b321a0b2
2 changed files with 83 additions and 32 deletions
+68 -28
View File
@@ -664,8 +664,8 @@
"source": "catalog created=2026-09-10"
},
"gemini": {
"name": "gemini-3.7-flash",
"source": "catalog created=2026-08-13"
"name": "gemini-3.8-flash",
"source": "catalog created=2026-09-02"
},
"gemma": {
"name": "gemma-4-31b-it",
@@ -680,8 +680,8 @@
"source": "catalog created=2026-09-04"
},
"grok": {
"name": "grok-4.5",
"source": "catalog created=2026-07-08"
"name": "grok-4.6",
"source": "catalog created=2026-08-12"
},
"inkling": {
"name": "inkling",
@@ -908,6 +908,26 @@
"x": -0.5482108382155133,
"y": 0.5760765801607652
},
{
"color": "#2563eb",
"family": "gpt",
"label": null,
"name": "gpt-oss-120b",
"provenance": {
"hle_score": null,
"hle_source_effort": null,
"hle_source_name": null,
"items": 12,
"protocol_id": "b5a600e549f8aaef4077c383d11a93d6b1f4395af26e0cc0f8f50d3b6f360996",
"readout": "rated categorical response",
"release_created": "2025-08-05",
"release_source": "catalog",
"run_id": "20260917T130652Z_b5a600e549f8",
"samples": 12
},
"x": -0.47891715945637514,
"y": 0.5996774327014757
},
{
"color": "#7e22ce",
"family": "qwen",
@@ -1111,7 +1131,7 @@
{
"color": "#2b2d42",
"family": "grok",
"label": null,
"label": "grok-4.6",
"name": "grok-4.6",
"provenance": {
"hle_score": null,
@@ -1408,26 +1428,6 @@
"x": -0.64,
"y": 0.64
},
{
"color": "#0ea5e9",
"family": "gemini",
"label": null,
"name": "gemini-2.5-flash-lite",
"provenance": {
"hle_score": null,
"hle_source_effort": null,
"hle_source_name": null,
"items": 12,
"protocol_id": "009e9bcb85966922ddafb0bb61fd0f750528edf0a13a21cf954b226476428210",
"readout": "rated categorical response",
"release_created": "2025-07-22",
"release_source": "catalog",
"run_id": "20260917T130434Z_009e9bcb8596",
"samples": 12
},
"x": -0.3863756613756613,
"y": 0.6708921222810112
},
{
"color": "#7e22ce",
"family": "qwen",
@@ -1571,7 +1571,7 @@
{
"color": "#2b2d42",
"family": "grok",
"label": "grok-4.5",
"label": null,
"name": "grok-4.5",
"provenance": {
"hle_score": null,
@@ -1851,7 +1851,7 @@
{
"color": "#0ea5e9",
"family": "gemini",
"label": null,
"label": "gemini-3.8-flash",
"name": "gemini-3.8-flash",
"provenance": {
"hle_score": null,
@@ -2128,6 +2128,26 @@
"x": -0.48767589171684006,
"y": 0.6482142857142857
},
{
"color": "#0ea5e9",
"family": "gemini",
"label": null,
"name": "gemini-2.5-flash",
"provenance": {
"hle_score": null,
"hle_source_effort": null,
"hle_source_name": null,
"items": 12,
"protocol_id": "a544b24e1874f8d1930042f65a61ea53299057a6299b0eb2ded110691c0d4edb",
"readout": "rated categorical response",
"release_created": "2025-06-17",
"release_source": "catalog",
"run_id": "20260917T130734Z_a544b24e1874",
"samples": 12
},
"x": -0.6298593073593075,
"y": 0.6495670432204361
},
{
"color": "#7e22ce",
"family": "qwen",
@@ -2168,6 +2188,26 @@
"x": -0.5966931216931217,
"y": 0.5338064713064713
},
{
"color": "#0ea5e9",
"family": "gemini",
"label": null,
"name": "gemini-2.5-flash-lite",
"provenance": {
"hle_score": null,
"hle_source_effort": null,
"hle_source_name": null,
"items": 12,
"protocol_id": "009e9bcb85966922ddafb0bb61fd0f750528edf0a13a21cf954b226476428210",
"readout": "rated categorical response",
"release_created": "2025-07-22",
"release_source": "catalog",
"run_id": "20260917T130434Z_009e9bcb8596",
"samples": 12
},
"x": -0.3863756613756613,
"y": 0.6708921222810112
},
{
"color": "#2563eb",
"family": "gpt",
@@ -2591,7 +2631,7 @@
{
"color": "#0ea5e9",
"family": "gemini",
"label": "gemini-3.7-flash",
"label": null,
"name": "gemini-3.7-flash",
"provenance": {
"hle_score": 0.478683966635774,
+15 -4
View File
@@ -475,13 +475,24 @@ def main() -> None:
name: datetime.fromtimestamp(saved_catalog[model_id]["created"], tz=timezone.utc).date().isoformat()
for name, model_id in LEGACY_CATALOG_IDS.items()
}
completed_by_name = {entry["display_key"].replace(" (rated)", ""): entry
for entry in cache["completed"].values()}
def release_created(name: str) -> str | None:
if name in metadata:
return metadata[name]["created"]
panel = completed_by_name.get(name)
if panel is not None and panel["model"] in saved_catalog:
return datetime.fromtimestamp(saved_catalog[panel["model"]]["created"], tz=timezone.utc).date().isoformat()
return legacy_release_dates.get(name)
model_labels: dict[str, str] = {}
label_sources: dict[str, str] = {}
for family, names in fams.items():
catalogued = [name for name in names if name in metadata]
if catalogued:
latest = max(catalogued, key=lambda name: metadata[name]["created"])
label_sources[family] = f"catalog created={metadata[latest]['created']}"
dated = [name for name in names if release_created(name) is not None]
if dated:
latest = max(dated, key=release_created)
label_sources[family] = f"catalog created={release_created(latest)}"
else:
latest = LEGACY_LATEST[family]
if latest not in names: