Spaces:
Running
Running
File size: 50,213 Bytes
c1f2130 bca888a 11542d9 6cc7b0b bca888a c1f2130 dd0b4fc c1f2130 2fcae3f c1f2130 8058fce c1f2130 8058fce c1f2130 415ac43 c1f2130 8058fce c1f2130 8058fce c1f2130 2fcae3f fe99ffa 2fcae3f 8058fce 2fcae3f c1f2130 8058fce c1f2130 8058fce c1f2130 8058fce c1f2130 8058fce c1f2130 8058fce c1f2130 8058fce c1f2130 8058fce c1f2130 8058fce c1f2130 8058fce fe99ffa 8058fce c1f2130 bca888a c1f2130 dd0b4fc c1f2130 bca888a c1f2130 415ac43 c1f2130 bca888a c1f2130 2fcae3f 431b0cc c1f2130 bca888a c1f2130 415ac43 c1f2130 415ac43 c1f2130 bca888a c1f2130 415ac43 c1f2130 2fcae3f c1f2130 bca888a c1f2130 bca888a c1f2130 415ac43 c1f2130 415ac43 c1f2130 2fcae3f c1f2130 415ac43 c1f2130 dd0b4fc c1f2130 dd0b4fc c1f2130 436ada0 dd0b4fc 436ada0 c1f2130 436ada0 c1f2130 fe99ffa c1f2130 fe99ffa c1f2130 fe99ffa bca888a 11542d9 bca888a 11542d9 2ed4959 11542d9 2ed4959 11542d9 2ed4959 11542d9 2ed4959 11542d9 2ed4959 11542d9 2ed4959 11542d9 415ac43 fe99ffa 415ac43 6cc7b0b bca888a fe99ffa bca888a c1f2130 415ac43 c1f2130 415ac43 c1f2130 415ac43 c1f2130 2fcae3f c1f2130 2fcae3f c1f2130 2fcae3f c1f2130 415ac43 c1f2130 bca888a c1f2130 2fcae3f c1f2130 2fcae3f c1f2130 2fcae3f c1f2130 415ac43 c1f2130 2ed4959 c1f2130 2ed4959 c1f2130 2fcae3f c1f2130 431b0cc 2fcae3f 431b0cc c1f2130 431b0cc 2fcae3f 431b0cc c1f2130 2fcae3f c1f2130 2fcae3f c1f2130 2fcae3f c1f2130 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 | import "server-only"
import { promises as fs } from "fs"
import path from "path"
import type {
BackendManifest,
BackendManifestStatus,
ComparisonIndex,
CorpusAggregates,
EvalHierarchy,
EvalcardsAnnotations,
HierarchyBenchmark,
HierarchyComposite,
HierarchyFamily,
HierarchyMetric,
HierarchySlice,
HierarchyTags,
PeerRanksMap,
RowAnnotations,
SignalSummaries,
} from "@/lib/backend-artifacts"
import type {
BenchmarkCard,
BenchmarkEvaluation,
CategoryType,
EvaluationResult,
MetricConfig,
ModelInfo,
SampleResult,
SourceData,
SourceMetadata,
} from "@/lib/benchmark-schema"
import { getCanonicalModelIdentity, getModelFamilyRouteId } from "@/lib/model-family"
// ---------------------------------------------------------------------------
// HuggingFace dataset base URL
// ---------------------------------------------------------------------------
const HF_DATASET = "evaleval/card_backend"
const HF_BASE = `https://huggingface.co/datasets/${HF_DATASET}/resolve/main`
// ---------------------------------------------------------------------------
// Local disk cache (populated by scripts/cache-hf-data.mjs during build)
// ---------------------------------------------------------------------------
// HF_DATA_LOCAL_DIR overrides the cache location so the JSON read path can be
// pointed at a sibling repo's pipeline output for parity testing against the
// DuckDB backend. Falls back to the cache populated by scripts/cache-hf-data.mjs.
const LOCAL_CACHE_DIR = process.env.HF_DATA_LOCAL_DIR?.trim()
? path.resolve(process.env.HF_DATA_LOCAL_DIR.trim())
: path.join(process.cwd(), ".cache", "hf-data")
async function readLocalCache<T>(relativePath: string): Promise<T | null> {
try {
const filePath = path.join(LOCAL_CACHE_DIR, relativePath)
const text = await fs.readFile(filePath, "utf8")
return JSON.parse(text) as T
} catch {
return null
}
}
// ---------------------------------------------------------------------------
// In-memory cache (always active to avoid HF rate limits)
// ---------------------------------------------------------------------------
interface CacheEntry {
data: unknown
ts: number
manifestSignature?: string
}
const cache = new Map<string, CacheEntry>()
const CACHE_TTL_MS =
process.env.HF_DATA_CACHE_TTL_MS != null
? Number.parseInt(process.env.HF_DATA_CACHE_TTL_MS, 10)
: process.env.NODE_ENV === "production"
? 60 * 1000
: 0
const MANIFEST_TTL_MS =
process.env.HF_MANIFEST_CACHE_TTL_MS != null
? Number.parseInt(process.env.HF_MANIFEST_CACHE_TTL_MS, 10)
: process.env.NODE_ENV === "production"
? 30 * 1000
: 0
let manifestSnapshotCache:
| {
remote: BackendManifest | null
local: BackendManifest | null
ts: number
}
| null = null
let localManifestPromise: Promise<BackendManifest | null> | null = null
let manifestRefreshPromise: Promise<void> | null = null
let refreshTargetManifestSignature: string | null = null
let refreshTargetFailed = false
let activeManifestSignature: string | null = null
let activeManifest: BackendManifest | null = null
const backgroundRefreshes = new Map<string, Promise<void>>()
const observedPaths = new Set<string>()
function isCanonicalCacheShape(relativePath: string, data: unknown) {
if (!data || typeof data !== "object") {
return false
}
const record = data as Record<string, unknown>
if (relativePath.startsWith("models/")) {
return record.hierarchy_by_category != null
}
if (relativePath.startsWith("evals/")) {
return Array.isArray(record.metrics)
}
if (relativePath === "eval-list.json") {
const evals = Array.isArray(record.evals) ? (record.evals as Array<Record<string, unknown>>) : []
return evals.length === 0 || typeof evals[0]?.benchmark_family_key === "string"
}
if (relativePath === "comparison-index.json") {
return record.evals != null && record.by_model != null
}
return true
}
function getManifestSignature(manifest: BackendManifest | null | undefined) {
if (!manifest) {
return null
}
return JSON.stringify({
generated_at: manifest.generated_at,
config_version: manifest.config_version,
skipped_configs: [...manifest.skipped_configs].sort(),
})
}
// HF_DATA_OFFLINE disables every network fetch, so the read path is fully
// served by LOCAL_CACHE_DIR. Used by the DuckDB parity setup so two servers
// reading the same on-disk artifacts cannot diverge mid-test via background
// refresh, and useful generally for offline development.
const OFFLINE = process.env.HF_DATA_OFFLINE === "1"
const DATA_BACKEND_VERSION = process.env.DATA_BACKEND?.trim().toLowerCase()
function useViewLayerBackend() {
return DATA_BACKEND_VERSION === "v2" || DATA_BACKEND_VERSION === "stage-j"
}
async function fetchSnapshotSidecars() {
return import("@/lib/sidecars")
}
async function fetchRemoteJson<T>(relativePath: string): Promise<T> {
if (OFFLINE) {
throw new Error(`HF_DATA_OFFLINE=1: refusing remote fetch for ${relativePath}`)
}
const url = `${HF_BASE}/${relativePath}`
let lastError: Error | null = null
for (let attempt = 0; attempt < 3; attempt++) {
if (attempt > 0) {
await new Promise((resolve) => setTimeout(resolve, 1000 * attempt))
}
try {
const res = await fetch(url, { cache: "no-store" })
if (res.status === 429) {
lastError = new Error(`HF rate limited (429) for ${url}`)
continue
}
if (!res.ok) {
throw new Error(`HF fetch failed: ${res.status} ${res.statusText} for ${url}`)
}
return (await res.json()) as T
} catch (err) {
lastError = err instanceof Error ? err : new Error(String(err))
if (!String(err).includes("429")) {
throw err
}
}
}
throw lastError ?? new Error(`HF fetch failed for ${url}`)
}
async function getLocalManifest() {
if (!localManifestPromise) {
localManifestPromise = readLocalCache<BackendManifest>("manifest.json")
}
const local = await localManifestPromise
if (!manifestSnapshotCache) {
manifestSnapshotCache = {
remote: null,
local,
ts: 0,
}
} else if (manifestSnapshotCache.local == null && local != null) {
manifestSnapshotCache.local = local
}
if (!activeManifestSignature) {
activeManifestSignature = getManifestSignature(local)
activeManifest = local
}
return local
}
function queueArtifactRefresh(
relativePath: string,
manifestSignature: string,
remoteManifest: BackendManifest | null
) {
if (backgroundRefreshes.has(relativePath)) {
return
}
if (backgroundRefreshes.size === 0 || refreshTargetManifestSignature !== manifestSignature) {
refreshTargetManifestSignature = manifestSignature
refreshTargetFailed = false
}
const refreshPromise = (async () => {
try {
const data = await fetchRemoteJson<unknown>(relativePath)
if (isCanonicalCacheShape(relativePath, data)) {
cache.set(relativePath, {
data,
ts: Date.now(),
manifestSignature,
})
}
} catch (err) {
refreshTargetFailed = true
console.warn(`[hf-data] Background refresh failed for ${relativePath}:`, err)
} finally {
backgroundRefreshes.delete(relativePath)
if (
backgroundRefreshes.size === 0 &&
!refreshTargetFailed &&
refreshTargetManifestSignature === manifestSignature
) {
activeManifestSignature = manifestSignature
activeManifest = remoteManifest
}
}
})()
backgroundRefreshes.set(relativePath, refreshPromise)
}
function queueObservedPathRefreshes(snapshot: {
remote: BackendManifest | null
local: BackendManifest | null
}) {
const remoteManifestSignature = getManifestSignature(snapshot.remote)
if (!remoteManifestSignature || remoteManifestSignature === activeManifestSignature) {
return
}
for (const relativePath of observedPaths) {
if (relativePath !== "manifest.json") {
queueArtifactRefresh(relativePath, remoteManifestSignature, snapshot.remote)
}
}
}
function queueManifestSnapshotRefresh() {
if (manifestRefreshPromise) {
return manifestRefreshPromise
}
manifestRefreshPromise = (async () => {
const local = await getLocalManifest()
const remote = await fetchRemoteJson<BackendManifest>("manifest.json").catch((err) => {
console.warn("[hf-data] Failed to fetch remote manifest:", err)
return null
})
manifestSnapshotCache = {
remote,
local,
ts: Date.now(),
}
const remoteManifestSignature = getManifestSignature(remote)
if (!activeManifestSignature && remoteManifestSignature) {
activeManifestSignature = remoteManifestSignature
activeManifest = remote
}
queueObservedPathRefreshes(manifestSnapshotCache)
})().finally(() => {
manifestRefreshPromise = null
})
return manifestRefreshPromise
}
async function getManifestSnapshot() {
const local = await getLocalManifest()
if (!manifestSnapshotCache) {
manifestSnapshotCache = {
remote: null,
local,
ts: 0,
}
}
if (
MANIFEST_TTL_MS === 0 ||
Date.now() - manifestSnapshotCache.ts >= MANIFEST_TTL_MS
) {
void queueManifestSnapshotRefresh()
}
return manifestSnapshotCache
}
function getCurrentManifestFromSnapshot(snapshot: {
remote: BackendManifest | null
local: BackendManifest | null
}) {
const remoteSignature = getManifestSignature(snapshot.remote)
if (remoteSignature && remoteSignature === activeManifestSignature) {
return snapshot.remote
}
return activeManifest ?? snapshot.local ?? snapshot.remote
}
async function fetchHFJson<T>(relativePath: string): Promise<T> {
if (relativePath === "manifest.json") {
const snapshot = await getManifestSnapshot()
if (snapshot.remote) {
return snapshot.remote as T
}
if (snapshot.local) {
return snapshot.local as T
}
throw new Error("HF manifest fetch failed and no local manifest cache is available")
}
const manifestSnapshot = await getManifestSnapshot()
const remoteManifestSignature = getManifestSignature(manifestSnapshot.remote)
const localManifestSignature = getManifestSignature(manifestSnapshot.local)
observedPaths.add(relativePath)
// 1. In-memory cache (hot)
const hit = cache.get(relativePath)
const validHotCache = hit ? isCanonicalCacheShape(relativePath, hit.data) : false
if (hit && !validHotCache) {
cache.delete(relativePath)
}
if (
hit &&
validHotCache &&
CACHE_TTL_MS > 0 &&
Date.now() - hit.ts < CACHE_TTL_MS &&
(!remoteManifestSignature || hit.manifestSignature === remoteManifestSignature)
) {
return hit.data as T
}
if (hit && validHotCache) {
if (remoteManifestSignature && hit.manifestSignature !== remoteManifestSignature) {
queueArtifactRefresh(relativePath, remoteManifestSignature, manifestSnapshot.remote)
}
return hit.data as T
}
const local = await readLocalCache<T>(relativePath)
const validLocalCache = local !== null && isCanonicalCacheShape(relativePath, local)
// 2. If the local cache was built from the same manifest, keep using it.
if (
validLocalCache &&
remoteManifestSignature &&
localManifestSignature &&
remoteManifestSignature === localManifestSignature
) {
cache.set(relativePath, {
data: local,
ts: Date.now(),
manifestSignature: remoteManifestSignature,
})
return local
}
// 3. Serve the local cache immediately and refresh in the background when the
// manifest indicates newer data exists.
if (validLocalCache) {
cache.set(relativePath, {
data: local,
ts: Date.now(),
manifestSignature: localManifestSignature ?? undefined,
})
if (remoteManifestSignature && remoteManifestSignature !== localManifestSignature) {
queueArtifactRefresh(relativePath, remoteManifestSignature, manifestSnapshot.remote)
}
return local
}
// 4. Fall back to a live fetch only when there is no usable stale cache.
try {
const data = await fetchRemoteJson<T>(relativePath)
cache.set(relativePath, {
data,
ts: Date.now(),
manifestSignature: remoteManifestSignature ?? undefined,
})
return data
} catch (err) {
if (hit && validHotCache) {
console.warn(`[hf-data] Using stale cache for ${relativePath} after live fetch failed`)
return hit.data as T
}
throw err
}
}
export async function fetchBackendManifestStatus(): Promise<BackendManifestStatus> {
if (useViewLayerBackend()) {
const manifest = await (await fetchSnapshotSidecars()).fetchManifest()
return {
currentManifest: manifest,
latestManifest: manifest,
currentManifestSignature: manifest.generated_at,
latestManifestSignature: manifest.generated_at,
updateAvailable: false,
refreshing: false,
pendingRefreshCount: 0,
}
}
const snapshot = await getManifestSnapshot()
const currentManifest = getCurrentManifestFromSnapshot(snapshot)
const currentManifestSignature = getManifestSignature(currentManifest)
const latestManifest = snapshot.remote ?? snapshot.local
const latestManifestSignature = getManifestSignature(latestManifest)
return {
currentManifest,
latestManifest,
currentManifestSignature,
latestManifestSignature,
updateAvailable: Boolean(
currentManifestSignature &&
latestManifestSignature &&
currentManifestSignature !== latestManifestSignature
),
refreshing: manifestRefreshPromise != null || backgroundRefreshes.size > 0,
pendingRefreshCount: backgroundRefreshes.size,
}
}
async function fetchHFJsonSafe<T>(relativePath: string): Promise<T | null> {
try {
return await fetchHFJson<T>(relativePath)
} catch (err) {
if (!String(err).includes("404")) {
console.warn(`[hf-data] Failed to fetch ${relativePath}:`, err)
}
return null
}
}
// ---------------------------------------------------------------------------
// HF dataset types (shapes of JSON files in the HF repo)
// ---------------------------------------------------------------------------
export interface HFModelCardEntry extends SignalSummaries {
model_family_id: string
model_route_id: string
model_family_name: string
developer: string
params_billions?: number | string | null
total_evaluations: number
benchmark_count: number
benchmark_family_count: number
categories_covered: string[]
last_updated: string
variants: Array<{
variant_key: string
variant_label: string
evaluation_count: number
raw_model_ids: string[]
last_updated?: string
}>
score_summary: {
count: number
min: number
max: number
avg?: number
average?: number
}
benchmark_names?: string[]
top_benchmark_scores?: Array<{
benchmark: string
benchmarkKey?: string
evaluation_name?: string
score: number
metric: string
lower_is_better?: boolean
}>
}
export interface HFEvalListEntry extends SignalSummaries {
eval_summary_id: string
benchmark: string
canonical_display_name?: string
benchmark_family_key: string
benchmark_family_name: string
benchmark_parent_key: string
benchmark_parent_name?: string
benchmark_leaf_key: string
benchmark_leaf_name: string
benchmark_component_key?: string | null
benchmark_component_name?: string | null
evaluation_name?: string
display_name: string
is_summary_score?: boolean
summary_score_for?: string | null
summary_score_for_name?: string | null
summary_eval_ids?: string[]
category: string
tags: {
domains: string[]
languages: string[]
tasks: string[]
}
models_count: number
metrics_count: number
subtasks_count?: number
metric_names: string[]
primary_metric_name: string
benchmark_card: BenchmarkCard | null
source_data?: SourceData
top_score: number
instance_data: {
available: boolean
url_count: number
sample_urls: string[]
models_with_loaded_instances: number
}
metrics: Array<{
metric_summary_id: string
metric_name: string
lower_is_better: boolean
models_count: number
top_score: number
}>
evalcards?: { annotations?: EvalcardsAnnotations }
}
export interface HFEvalModelResult {
model_id: string
model_route_id: string
model_name: string
developer: string
raw_model_id?: string
score: number
evaluation_id?: string
retrieved_timestamp?: string
source_record_url?: string
// The pipeline copies the parent record's provenance onto every hierarchy
// model_result row (commit 9090cc5, 2026-04-26). Required.
source_metadata: SourceMetadata
source_data?: SourceData | string[]
detailed_evaluation_results?: string | null
detailed_evaluation_results_meta?: unknown
instance_level_data?: unknown
passthrough_top_level_fields?: unknown
evalcards?: { annotations?: RowAnnotations }
}
export interface HFEvalMetric {
metric_summary_id: string
legacy_eval_summary_id?: string
evaluation_name?: string
metric_name: string
metric_key: string
display_name?: string
canonical_display_name?: string
metric_config?: MetricConfig | Record<string, unknown>
lower_is_better: boolean
model_results: HFEvalModelResult[]
}
export interface HFEvalDetail extends SignalSummaries {
eval_summary_id: string
benchmark: string
canonical_display_name?: string
benchmark_family_key: string
benchmark_leaf_key: string
benchmark_leaf_name: string
benchmark_parent_key?: string
benchmark_parent_name?: string
category: string
source_data: SourceData
benchmark_card: BenchmarkCard | null
metrics: HFEvalMetric[]
subtasks: unknown[]
evalcards?: { annotations?: EvalcardsAnnotations }
}
export interface HFModelDetail extends SignalSummaries {
model_info: ModelInfo & {
family_id?: string
family_slug?: string
variant_key?: string
variant_label?: string
model_route_id?: string
}
model_family_id: string
model_route_id: string
model_family_name?: string
raw_model_ids?: string[]
last_updated?: string
hierarchy_by_category: Record<string, HFModelHierarchyNode[]>
evaluations_by_category?: Record<string, HFModelEvaluation[]>
total_evaluations: number
categories_covered: string[]
variants: Array<{
variant_key: string
variant_label: string
evaluation_count: number
raw_model_ids: string[]
}>
}
/** Evaluation entry inside an HF model detail file */
export interface HFModelEvaluation {
schema_version?: string
evaluation_id: string
retrieved_timestamp: string
benchmark?: string
source_data?: SourceData | string[]
source_metadata?: SourceMetadata
eval_library?: { name: string; version?: string; additional_details?: Record<string, unknown> }
model_info?: ModelInfo
generation_config?: BenchmarkEvaluation["generation_config"]
evaluation_results: EvaluationResult[]
source_record_url?: string
detailed_evaluation_results?: string | null
detailed_evaluation_results_meta?: unknown
instance_level_data?: unknown
benchmark_card?: BenchmarkCard | null
passthrough_top_level_fields?: unknown
}
export interface HFModelHierarchyMetric {
metric_summary_id: string
legacy_eval_summary_id?: string
evaluation_name: string
display_name: string
canonical_display_name?: string
benchmark_leaf_key: string
benchmark_leaf_name: string
slice_key?: string | null
slice_name?: string | null
lower_is_better: boolean
metric_name: string
metric_id: string
metric_key: string
metric_source?: string
metric_config: MetricConfig
models_count: number
top_score?: number
model_results: HFEvalModelResult[]
}
export interface HFModelHierarchyNode {
eval_summary_id: string
benchmark: string
canonical_display_name?: string
benchmark_family_key: string
benchmark_family_name: string
benchmark_parent_key: string
benchmark_parent_name: string
benchmark_leaf_key: string
benchmark_leaf_name: string
benchmark_component_key?: string | null
benchmark_component_name?: string | null
evaluation_name: string
display_name: string
is_summary_score: boolean
category: string
source_data: SourceData
benchmark_card: BenchmarkCard | null
tags: {
domains: string[]
languages: string[]
tasks: string[]
}
subtasks_count: number
metrics_count: number
metric_names: string[]
primary_metric_name: string
metrics: HFModelHierarchyMetric[]
subtasks: HFModelHierarchyNode[]
top_score?: number
summary_eval_ids?: string[]
instance_data?: {
available: boolean
url_count: number
sample_urls: string[]
models_with_loaded_instances: number
}
}
export type HFModelHierarchySubtask = Partial<Omit<HFModelHierarchyNode, "subtasks">> & {
subtask_key?: string
subtask_name?: string
canonical_display_name?: string
metrics?: HFModelHierarchyMetric[]
subtasks?: HFModelHierarchySubtask[]
}
export interface HFDeveloperEntry {
developer: string
model_count: number
}
export interface HFDeveloperDetail {
developer: string
models: HFModelCardEntry[]
}
function normalizeSetupAliasQualifier(value: string | null | undefined) {
return value?.trim().toLowerCase().replace(/[_\s]+/g, "-") ?? ""
}
function isSetupAliasQualifier(value: string | null | undefined) {
const normalized = normalizeSetupAliasQualifier(value)
return (
normalized === "prompt" ||
normalized === "fc" ||
normalized === "function-calling" ||
normalized.startsWith("thinking")
)
}
function getLatestTimestamp(a?: string, b?: string) {
if (!a) return b
if (!b) return a
const aTime = new Date(a).getTime()
const bTime = new Date(b).getTime()
if (!Number.isFinite(aTime)) return b
if (!Number.isFinite(bTime)) return a
return bTime > aTime ? b : a
}
function sortNormalizedModelCardVariants(a: HFModelCardEntry["variants"][number], b: HFModelCardEntry["variants"][number]) {
const aIsDefault = a.variant_key === "default"
const bIsDefault = b.variant_key === "default"
if (aIsDefault !== bIsDefault) {
return aIsDefault ? -1 : 1
}
const aTime = a.last_updated ? new Date(a.last_updated).getTime() : Number.NEGATIVE_INFINITY
const bTime = b.last_updated ? new Date(b.last_updated).getTime() : Number.NEGATIVE_INFINITY
if (aTime !== bTime) {
return bTime - aTime
}
return a.variant_label.localeCompare(b.variant_label)
}
function normalizeSingleModelCardEntry(entry: HFModelCardEntry): HFModelCardEntry {
const familyIdentity = getCanonicalModelIdentity({
id: entry.model_family_id,
name: entry.model_family_name,
})
const variantsByKey = new Map<string, HFModelCardEntry["variants"][number]>()
for (const variant of entry.variants ?? []) {
let normalizedVariantKey = variant.variant_key
let normalizedVariantLabel = variant.variant_label
if (variant.variant_key === "base") {
normalizedVariantKey = "default"
normalizedVariantLabel = "Default"
} else if (variant.variant_key !== "default") {
const syntheticIdentity = getCanonicalModelIdentity({
id: `${familyIdentity.familyId}-${variant.variant_key}`,
name: `${familyIdentity.familyId}-${variant.variant_key}`,
})
if (syntheticIdentity.versionDate && isSetupAliasQualifier(syntheticIdentity.versionQualifier)) {
normalizedVariantKey = syntheticIdentity.versionDate
normalizedVariantLabel = syntheticIdentity.versionDate
} else {
normalizedVariantKey = syntheticIdentity.variantKey
normalizedVariantLabel = syntheticIdentity.variantLabel
}
}
const existing = variantsByKey.get(normalizedVariantKey)
if (existing) {
existing.evaluation_count += variant.evaluation_count
existing.last_updated = getLatestTimestamp(existing.last_updated, variant.last_updated)
existing.raw_model_ids = Array.from(
new Set([...(existing.raw_model_ids ?? []), ...(variant.raw_model_ids ?? [])])
).sort((a, b) => a.localeCompare(b))
continue
}
variantsByKey.set(normalizedVariantKey, {
...variant,
variant_key: normalizedVariantKey,
variant_label: normalizedVariantLabel,
raw_model_ids: [...(variant.raw_model_ids ?? [])].sort((a, b) => a.localeCompare(b)),
})
}
const normalizedVariants = Array.from(variantsByKey.values()).sort(sortNormalizedModelCardVariants)
const normalizedTotalEvaluations =
normalizedVariants.length > 0
? normalizedVariants.reduce((sum, variant) => sum + variant.evaluation_count, 0)
: entry.total_evaluations
return {
...entry,
model_family_id: familyIdentity.familyId,
model_route_id: getModelFamilyRouteId(familyIdentity.familyId),
model_family_name: familyIdentity.familyName,
total_evaluations: normalizedTotalEvaluations,
variants: normalizedVariants,
}
}
function normalizeModelCardEntries(entries: HFModelCardEntry[]) {
return entries.map(normalizeSingleModelCardEntry)
}
// ---------------------------------------------------------------------------
// Public data fetchers
// ---------------------------------------------------------------------------
export async function fetchModelCardsList(): Promise<HFModelCardEntry[]> {
const data = await fetchHFJson<HFModelCardEntry[]>("model-cards.json")
return Array.isArray(data) ? normalizeModelCardEntries(data) : []
}
export async function fetchModelCardsListLite(): Promise<HFModelCardEntry[]> {
const data = await fetchHFJsonSafe<HFModelCardEntry[]>("model-cards-lite.json")
if (Array.isArray(data)) {
return normalizeModelCardEntries(data)
}
return fetchModelCardsList()
}
export async function fetchEvalList(): Promise<{ evals: HFEvalListEntry[] }> {
return fetchHFJson<{ evals: HFEvalListEntry[] }>("eval-list.json")
}
export async function fetchEvalListLite(): Promise<{ evals: HFEvalListEntry[] }> {
const data = await fetchHFJsonSafe<{ evals: HFEvalListEntry[] }>("eval-list-lite.json")
if (data && Array.isArray(data.evals)) {
return data
}
return fetchEvalList()
}
export async function fetchDevelopersList(): Promise<HFDeveloperEntry[]> {
const data = await fetchHFJson<HFDeveloperEntry[]>("developers.json")
return Array.isArray(data) ? data : []
}
export async function fetchBenchmarkMetadataMap(): Promise<Record<string, BenchmarkCard>> {
if (useViewLayerBackend()) {
return (await import("@/lib/view-data")).getBenchmarkMetadataMap()
}
return fetchHFJson<Record<string, BenchmarkCard>>("benchmark-metadata.json")
}
export async function fetchBackendManifest(): Promise<BackendManifest> {
if (useViewLayerBackend()) {
return (await fetchSnapshotSidecars()).fetchManifest()
}
return fetchHFJson<BackendManifest>("manifest.json")
}
export async function fetchEvalHierarchy(): Promise<EvalHierarchy> {
if (useViewLayerBackend()) {
return adaptEvalHierarchy(await (await fetchSnapshotSidecars()).fetchHierarchy())
}
const raw = await fetchHFJson<EvalHierarchy>("eval-hierarchy.json")
return adaptEvalHierarchy(raw)
}
/**
* The upstream pipeline emits multiple shapes over its lifetime; this
* adapter promotes whichever form arrives into the legacy
* `families[].composites[]` / `families[].standalone_benchmarks[]`
* tree the existing UI walks, plus a complete `stats` block.
*
* Three shapes handled:
* 1. Legacy nested (`families[].composites[]`/`standalone_benchmarks[]`) —
* passed through.
* 2. Mid-life flat 2-level (`families[].leaves[]`) — leaves promoted to
* `standalone_benchmarks[]`.
* 3. New composite/family/slice taxonomy (top-level `composites[]` +
* flat `families[]` index) — composites' benchmarks are bucketed
* back under per-family records, with the composite slug as the
* family key when no curated multi-benchmark family applies.
*/
/**
* Validate-and-passthrough for the v3 hierarchy shape (per
* /Users/jchim/projects/evaleval/notes/hierarchy-alignment.md §5.1).
* The producer's `write_hierarchy()` emits family-rooted trees
* directly; the adapter no longer synthesises legacy shapes.
*
* Behaviour:
* - v3 detection via `schema_version === "v3.hierarchy.1"` —
* pass through unchanged.
* - Older snapshot lacking schema_version: log a warning and
* pass through. Consumers may render empty for missing fields
* but won't crash.
*
* Step 4 deleted the legacy `adaptCompositeShape()` synthesis
* (~400 lines of bucketing logic that built families[].composites[]
* from the old top-level composites[]). The producer now does that
* grouping at write time using `canonical_composites.family_id`.
*/
export function adaptEvalHierarchy(raw: EvalHierarchy): EvalHierarchy {
if (!raw) {
return { families: [] }
}
if (raw.schema_version && !raw.schema_version.startsWith("v3.hierarchy.")) {
console.warn(
`adaptEvalHierarchy: unexpected schema_version=${JSON.stringify(raw.schema_version)}; ` +
`expected v3.hierarchy.*. Frontend may render incompletely.`,
)
}
return raw
}
export async function fetchComparisonIndex(): Promise<ComparisonIndex> {
if (useViewLayerBackend()) {
return (await fetchSnapshotSidecars()).fetchComparisonIndex()
}
return fetchHFJson<ComparisonIndex>("comparison-index.json")
}
/**
* Per-(eval, model) primary-metric peer ranks. v2 reads the wrapped
* sidecar from the pinned snapshot; legacy reads the bare-map file
* historically published unversioned at the dataset root.
*/
export async function fetchPeerRanks(): Promise<PeerRanksMap> {
if (useViewLayerBackend()) {
return (await fetchSnapshotSidecars()).fetchPeerRanks()
}
return (await fetchHFJsonSafe<PeerRanksMap>("peer-ranks.json")) ?? {}
}
export async function fetchCorpusAggregates(): Promise<CorpusAggregates | null> {
if (useViewLayerBackend()) {
return (await fetchSnapshotSidecars()).fetchHeadline()
}
return fetchHFJsonSafe<CorpusAggregates>("corpus-aggregates.json")
}
export async function fetchModelDetail(slug: string): Promise<HFModelDetail | null> {
return fetchHFJsonSafe<HFModelDetail>(`models/${slug}.json`)
}
export async function fetchEvalDetail(slug: string): Promise<HFEvalDetail | null> {
return fetchHFJsonSafe<HFEvalDetail>(`evals/${slug}.json`)
}
export async function fetchDeveloperDetail(slug: string): Promise<HFDeveloperDetail | null> {
return fetchHFJsonSafe<HFDeveloperDetail>(`developers/${slug}.json`)
}
// ---------------------------------------------------------------------------
// Instance-level data fetching
// ---------------------------------------------------------------------------
/**
* Fetches instance-level sample data from a JSONL URL.
* Returns up to `limit` samples. If limit is 0 or undefined, returns all.
*/
export async function fetchInstanceLevelData(
url: string,
limit?: number
): Promise<SampleResult[]> {
try {
const res = await fetch(url)
if (!res.ok) return []
const text = await res.text()
const lines = text.split("\n").filter((line) => line.trim())
const maxLines = limit && limit > 0 ? Math.min(limit, lines.length) : lines.length
const parsed: unknown[] = []
for (let i = 0; i < maxLines; i++) {
try {
parsed.push(JSON.parse(lines[i]))
} catch {
// skip malformed line
}
}
// Reuse the same parser that handles the rich instance_examples format
return parseInstanceLevelData({ instance_examples: parsed })
} catch (err) {
console.warn("[hf-data] Failed to fetch instance-level data:", err)
return []
}
}
// ---------------------------------------------------------------------------
// Converters: HF shapes → app domain types
// ---------------------------------------------------------------------------
/**
* Parse instance_level_data from HF into SampleResult[].
*
* HF shape is an object:
* { interaction_type, instance_count, source_url, instance_examples: [...] }
*
* Each instance_example has: sample_id, input ({raw, reference}), output,
* messages (multi-turn), evaluation ({score, is_correct}), metadata, etc.
*/
// eslint-disable-next-line @typescript-eslint/no-explicit-any
function parseInstanceLevelData(data: unknown): SampleResult[] {
if (!data || typeof data !== "object") return []
// Extract the examples array — it lives under instance_examples
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const obj = data as Record<string, any>
const examples: unknown[] = Array.isArray(obj.instance_examples)
? obj.instance_examples
: Array.isArray(data)
? data
: []
if (examples.length === 0) return []
return examples
.map((raw: any, i: number) => {
if (!raw || typeof raw !== "object") return null
// Build a readable input string from the instance data
let input = ""
if (typeof raw.input === "string") {
input = raw.input
} else if (raw.input?.raw) {
input = String(raw.input.raw)
} else if (raw.prompt) {
input = raw.prompt
} else if (raw.question) {
input = raw.question
} else if (raw.doc?.question) {
input = raw.doc.question
} else if (raw.doc) {
input = JSON.stringify(raw.doc).slice(0, 500)
}
// Build a ground truth string
let groundTruth: string | undefined
if (raw.input?.reference) {
groundTruth = Array.isArray(raw.input.reference)
? raw.input.reference.join(", ")
: String(raw.input.reference)
} else if (raw.ground_truth != null) {
groundTruth = String(raw.ground_truth)
} else if (raw.target != null) {
groundTruth = String(raw.target)
} else if (raw.gold != null) {
groundTruth = String(raw.gold)
} else if (raw.doc?.answer != null) {
groundTruth = String(raw.doc.answer)
}
// Build a response string from output, messages, or answer_attribution
let response = ""
if (raw.output != null) {
response = typeof raw.output === "string" ? raw.output : JSON.stringify(raw.output)
} else if (raw.response) {
response = raw.response
} else if (raw.model_output) {
response = raw.model_output
} else if (Array.isArray(raw.answer_attribution) && raw.answer_attribution.length > 0) {
const attr = raw.answer_attribution[raw.answer_attribution.length - 1]
response = attr.extracted_value ?? ""
} else if (Array.isArray(raw.messages) && raw.messages.length > 0) {
// Use the last assistant message as the response
const lastAssistant = [...raw.messages]
.reverse()
.find((m: any) => m.role === "assistant")
if (lastAssistant) {
response = typeof lastAssistant.content === "string"
? lastAssistant.content
: JSON.stringify(lastAssistant.content)
}
} else if (raw.filtered_resps?.[0]?.[0]) {
response = raw.filtered_resps[0][0]
} else if (raw.resps?.[0]?.[0]) {
response = raw.resps[0][0]
}
// Determine correctness
const isCorrect =
raw.evaluation?.is_correct ??
raw.is_correct ??
(raw.metrics?.exact_match === 1 ? true :
raw.metrics?.exact_match === 0 ? false : undefined)
// Build metadata combining evaluation results and any extra metadata
const metadata: Record<string, any> = {}
if (raw.evaluation && typeof raw.evaluation === "object") {
Object.assign(metadata, raw.evaluation)
}
if (raw.performance && typeof raw.performance === "object") {
Object.assign(metadata, raw.performance)
}
if (raw.metadata && typeof raw.metadata === "object") {
Object.assign(metadata, raw.metadata)
}
if (raw.metrics && typeof raw.metrics === "object") {
Object.assign(metadata, raw.metrics)
}
return {
sample_id: raw.sample_id ?? raw.doc_id ?? raw.id ?? String(i),
input,
ground_truth: groundTruth,
response,
choices: raw.choices ?? raw.doc?.choices ?? undefined,
is_correct: isCorrect,
metadata: Object.keys(metadata).length > 0 ? metadata : undefined,
} as SampleResult
})
.filter((s): s is SampleResult => s !== null)
}
function normalizeModelIdForLookup(value: string | undefined | null) {
return value?.trim().toLowerCase() ?? ""
}
function toComparableTimestamp(timestamp: string | undefined) {
if (!timestamp) {
return Number.NEGATIVE_INFINITY
}
const numericTimestamp = Number.parseFloat(timestamp)
if (Number.isFinite(numericTimestamp)) {
return numericTimestamp
}
const parsedTimestamp = new Date(timestamp).getTime()
return Number.isFinite(parsedTimestamp) ? parsedTimestamp : Number.NEGATIVE_INFINITY
}
function buildVariantLookup(detail: HFModelDetail) {
const variantLookup = new Map<string, { variantKey: string; variantLabel: string }>()
for (const variant of detail.variants ?? []) {
for (const rawModelId of variant.raw_model_ids ?? []) {
const normalized = normalizeModelIdForLookup(rawModelId)
if (normalized) {
variantLookup.set(normalized, {
variantKey: variant.variant_key,
variantLabel: variant.variant_label,
})
}
}
}
return variantLookup
}
function resolveVariantMeta(
detail: HFModelDetail,
variantLookup: Map<string, { variantKey: string; variantLabel: string }>,
result: HFEvalModelResult
) {
const candidates = [result.raw_model_id, result.model_id]
.map((value) => normalizeModelIdForLookup(value))
.filter(Boolean)
for (const candidate of candidates) {
const matched = variantLookup.get(candidate)
if (matched) {
return matched
}
}
if ((detail.variants?.length ?? 0) === 1) {
return {
variantKey: detail.variants[0].variant_key,
variantLabel: detail.variants[0].variant_label,
}
}
return {
variantKey: candidates[0] || detail.model_info.variant_key || "default",
variantLabel: detail.model_info.variant_label || "Default",
}
}
function belongsToModelFamily(
detail: HFModelDetail,
result: HFEvalModelResult,
rawModelIds: Set<string>
) {
const routeId = normalizeModelIdForLookup(result.model_route_id)
if (routeId && routeId === normalizeModelIdForLookup(detail.model_route_id)) {
return true
}
const rawModelId = normalizeModelIdForLookup(result.raw_model_id)
if (rawModelId && rawModelIds.has(rawModelId)) {
return true
}
const modelId = normalizeModelIdForLookup(result.model_id)
if (modelId && rawModelIds.has(modelId)) {
return true
}
return false
}
function buildModelInfoForVariant(
detail: HFModelDetail,
result: HFEvalModelResult,
variantMeta: { variantKey: string; variantLabel: string }
): ModelInfo {
const modelId = result.raw_model_id ?? result.model_id ?? detail.model_info.id
const modelName = result.model_name || detail.model_family_name || detail.model_info.name
const variantLabel = variantMeta.variantLabel && variantMeta.variantLabel !== "Default"
? variantMeta.variantLabel
: undefined
return {
...detail.model_info,
id: modelId,
name: modelName,
developer: result.developer || detail.model_info.developer,
model_version: variantLabel,
additional_details: {
...detail.model_info.additional_details,
raw_model_id: result.raw_model_id ?? result.model_id,
},
}
}
function getCanonicalInstanceResultsUrl(value: unknown) {
if (typeof value !== "string") {
return undefined
}
return value.includes("/datasets/evaleval/card_backend/") && value.includes("/instances/")
? value
: undefined
}
function getNodeSubtaskKey(node: HFModelHierarchyNode | HFModelHierarchySubtask) {
return "subtask_key" in node ? node.subtask_key : undefined
}
function getNodeSubtaskName(node: HFModelHierarchyNode | HFModelHierarchySubtask) {
return "subtask_name" in node ? node.subtask_name : undefined
}
interface FlattenHierarchyContext {
eval_summary_id?: string
benchmark?: string
display_name?: string
canonical_display_name?: string
sourceData: SourceData
benchmark_family_key?: string
benchmark_family_name?: string
benchmark_parent_key?: string
benchmark_parent_name?: string
benchmark_leaf_key?: string
benchmark_leaf_name?: string
}
function buildFlattenHierarchyContext(
node: HFModelHierarchyNode | HFModelHierarchySubtask,
inheritedContext?: FlattenHierarchyContext
): FlattenHierarchyContext {
const benchmark = node.benchmark ?? inheritedContext?.benchmark
const benchmarkFamilyName = node.benchmark_family_name ?? inheritedContext?.benchmark_family_name
const displayName =
node.display_name ??
getNodeSubtaskName(node) ??
node.benchmark_leaf_name ??
inheritedContext?.benchmark_leaf_name ??
benchmarkFamilyName ??
benchmark ??
"Unknown Benchmark"
const canonicalDisplayName =
node.canonical_display_name ??
inheritedContext?.canonical_display_name ??
displayName
const sourceData =
node.source_data ??
inheritedContext?.sourceData ?? {
dataset_name: benchmark ?? displayName,
}
return {
eval_summary_id: node.eval_summary_id ?? inheritedContext?.eval_summary_id,
benchmark,
display_name: displayName,
canonical_display_name: canonicalDisplayName,
sourceData,
benchmark_family_key: node.benchmark_family_key ?? inheritedContext?.benchmark_family_key,
benchmark_family_name: benchmarkFamilyName,
benchmark_parent_key: node.benchmark_parent_key ?? inheritedContext?.benchmark_parent_key,
benchmark_parent_name: node.benchmark_parent_name ?? inheritedContext?.benchmark_parent_name,
benchmark_leaf_key: node.benchmark_leaf_key ?? inheritedContext?.benchmark_leaf_key,
benchmark_leaf_name: node.benchmark_leaf_name ?? inheritedContext?.benchmark_leaf_name,
}
}
function flattenHierarchyNode(
detail: HFModelDetail,
node: HFModelHierarchyNode | HFModelHierarchySubtask,
category: CategoryType,
rawModelIds: Set<string>,
variantLookup: Map<string, { variantKey: string; variantLabel: string }>,
inheritedContext?: FlattenHierarchyContext
): BenchmarkEvaluation[] {
const evaluations: BenchmarkEvaluation[] = []
const context = buildFlattenHierarchyContext(node, inheritedContext)
const sourceData = context.sourceData
for (const metric of node.metrics ?? []) {
const relevantResults = (metric.model_results ?? []).filter((result) =>
belongsToModelFamily(detail, result, rawModelIds)
)
if (relevantResults.length === 0) {
continue
}
const resultsByVariant = new Map<
string,
{
modelInfo: ModelInfo
evaluationResults: EvaluationResult[]
inlineSamples?: SampleResult[]
latestTimestamp: string
sourceMetadata: SourceMetadata
}
>()
for (const result of relevantResults) {
// Pipeline contract (commit 9090cc5): every model_result row carries
// source_metadata. Fail loud if a stale dataset breaks the contract —
// the UI dereferences source_metadata.* unguarded.
if (!result.source_metadata) {
throw new Error(
`Pipeline contract broken: missing source_metadata on model_result ` +
`(model_family=${detail.model_family_id} metric=${metric.metric_summary_id} eval=${result.evaluation_id})`
)
}
const variantMeta = resolveVariantMeta(detail, variantLookup, result)
const variantKey = variantMeta.variantKey || "default"
const modelInfo = buildModelInfoForVariant(detail, result, variantMeta)
const inlineSamples = parseInstanceLevelData(result.instance_level_data)
const evaluationResult: EvaluationResult = {
evaluation_name: metric.metric_name || metric.evaluation_name || metric.display_name,
display_name: metric.display_name || metric.metric_name || metric.evaluation_name,
canonical_display_name:
metric.canonical_display_name ||
metric.display_name ||
`${context.benchmark ?? context.display_name ?? "Benchmark"} / ${metric.metric_name}`,
metric_summary_id: metric.metric_summary_id,
metric_key: metric.metric_key,
evaluation_timestamp: result.retrieved_timestamp ?? detail.last_updated ?? "",
source_data: sourceData,
metric_config: metric.metric_config,
score_details: {
score: result.score,
},
detailed_evaluation_results_url: getCanonicalInstanceResultsUrl(
result.detailed_evaluation_results
),
evalcards: result.evalcards,
}
const existing = resultsByVariant.get(variantKey)
if (!existing) {
resultsByVariant.set(variantKey, {
modelInfo,
evaluationResults: [evaluationResult],
inlineSamples: inlineSamples.length > 0 ? inlineSamples : undefined,
latestTimestamp: result.retrieved_timestamp ?? detail.last_updated ?? "",
sourceMetadata: result.source_metadata,
})
continue
}
existing.evaluationResults.push(evaluationResult)
if ((!existing.inlineSamples || existing.inlineSamples.length === 0) && inlineSamples.length > 0) {
existing.inlineSamples = inlineSamples
}
if (
toComparableTimestamp(result.retrieved_timestamp) >=
toComparableTimestamp(existing.latestTimestamp)
) {
// When multiple submissions land in the same variant bucket, prefer
// provenance from the freshest one.
existing.latestTimestamp = result.retrieved_timestamp ?? existing.latestTimestamp
existing.sourceMetadata = result.source_metadata
}
}
for (const [variantKey, variantGroup] of resultsByVariant.entries()) {
const sliceKey = metric.slice_key ?? getNodeSubtaskKey(node) ?? undefined
const sliceName = metric.slice_name ?? getNodeSubtaskName(node) ?? undefined
evaluations.push({
schema_version: "0.2.2",
eval_summary_id: context.eval_summary_id,
evaluation_id: `${metric.metric_summary_id}__${variantKey}`,
retrieved_timestamp: variantGroup.latestTimestamp,
benchmark: context.benchmark,
display_name:
node.display_name ??
getNodeSubtaskName(node) ??
metric.slice_name ??
context.display_name ??
context.benchmark_leaf_name ??
context.benchmark,
canonical_display_name:
node.canonical_display_name ??
(metric.slice_name && (context.benchmark_parent_name ?? context.benchmark)
? `${context.benchmark_parent_name ?? context.benchmark} / ${metric.slice_name}`
: context.canonical_display_name ?? context.benchmark),
category,
family_id: context.benchmark_family_key,
benchmark_family_name: context.benchmark_family_name,
parent_benchmark_id: context.benchmark_parent_key,
benchmark_parent_name: context.benchmark_parent_name,
benchmark_leaf_name: metric.benchmark_leaf_name ?? context.benchmark_leaf_name,
slice_key: sliceKey,
slice_name: sliceName,
source_data: sourceData,
source_metadata: variantGroup.sourceMetadata,
model_info: variantGroup.modelInfo,
evaluation_results: variantGroup.evaluationResults,
detailed_evaluation_results_per_samples:
variantGroup.inlineSamples && variantGroup.inlineSamples.length > 0
? variantGroup.inlineSamples
: undefined,
})
}
}
for (const subtask of node.subtasks ?? []) {
evaluations.push(
...flattenHierarchyNode(
detail,
subtask,
category,
rawModelIds,
variantLookup,
context
)
)
}
return evaluations
}
/**
* Flatten hierarchy_by_category from an HF model detail file into BenchmarkEvaluation[]
* while preserving backend-declared categories, grouping keys, and variant boundaries.
*/
export function flattenModelEvaluations(detail: HFModelDetail): BenchmarkEvaluation[] {
const evaluations: BenchmarkEvaluation[] = []
const rawModelIds = new Set(
[
...(detail.raw_model_ids ?? []),
...((detail.variants ?? []).flatMap((variant) => variant.raw_model_ids ?? [])),
detail.model_info.id,
detail.model_family_id,
]
.map((value) => normalizeModelIdForLookup(value))
.filter(Boolean)
)
const variantLookup = buildVariantLookup(detail)
for (const [categoryKey, nodes] of Object.entries(detail.hierarchy_by_category ?? {})) {
const mappedCategory = mapHFCategories([categoryKey])[0]
for (const node of nodes) {
evaluations.push(
...flattenHierarchyNode(
detail,
node,
mappedCategory,
rawModelIds,
variantLookup
)
)
}
}
return evaluations
}
/**
* Map pipeline category labels to frontend CategoryType.
*/
// Every category key emitted by the pipeline (verified against production
// dataset 2026-04-27, 9 distinct keys total). Values for the 3 added keys
// (coding, instruction_following, language_understanding) match what the
// previous regex fallback returned for them, preserving prior labelling.
// Note: `coding` maps to General because "coding" does not contain the
// substring "code" — see lib/benchmark-schema.ts inferCategoryFromBenchmark.
const PIPELINE_CATEGORY_MAP: Record<string, CategoryType> = {
agentic: "Agentic",
reasoning: "Reasoning",
general: "General",
safety: "Safety",
knowledge: "Knowledge",
other: "General",
coding: "General",
instruction_following: "General",
language_understanding: "General",
}
export function mapHFCategories(categories: string[]): CategoryType[] {
const mapped: CategoryType[] = []
for (const c of categories) {
if (!c) continue
const cat = PIPELINE_CATEGORY_MAP[c.toLowerCase()] ?? "General"
if (!mapped.includes(cat)) mapped.push(cat)
}
return mapped.length > 0 ? mapped : ["General"]
}
|