ScalingIntelligence/KernelBench — GitHub trend

◂ BACK TO LEADERBOARD

FINDING #3590 · UNIT ID 878197542

ScalingIntelligence/KernelBench

KernelBench: Can LLMs Write GPU Kernels? - Benchmark + Toolkit with Torch -> CUDA (+ more DSLs)

[ JUPYTER NOTEBOOK ][ ORG ][ VERIFIED ][ GITHUB ↗ ]

SURPRISE SCORE

0.00

Score Breakdown

SURPRISE0.00272

ENGAGEMENT0.20

FRESHNESS1.00

SCORE = SURPRISE × ENGAGEMENT^0.7 × FRESHNESS × VISIBILITY × CONFIDENCE

SURPRISE = WINDOW STARS / DAYS / (AUDIENCE + FLOOR)

1% OF STARS IN ARCHIVE

Growth Telemetry

VELOCITY /D

3.33

ACCEL

-1.00

RETENTION

50.0%

PEAK 2026-07-13 · FORK-RETENTION 0.0% · 10 STARS / WINDOW

Author Audience

AUDIENCE

1,187

FOLLOWERS

330

OWNER ★

2,370

Engagement Signals

FORKS

178

ISSUE AUTH

PR AUTH

UNIQUE STARGAZERS 10 / 10 (DIVERSITY 1.00)

Why This Is A Finding

ScalingIntelligence/KernelBench собрал 10 звёзд за окно, тогда как у автора всего 330 подписчиков — эффективная аудитория ≈ 1,187. Это даёт surprise-индекс 0.00272 (звёзды относительно охвата автора, а не в абсолюте). Удержание форков 0.0% и 0 внешних контрибьюторов отделяют реальный инструмент от разовой вспышки. Акселерация отрицательная — внимание остывает после пика.

Related Findings

RANKS ABOVE 29% OF 5024 FINDINGS

#704lit26/finvizfinance[ JUPYTER NOTEBOOK ]0.0141 #792CalvinXKY/InfraTech[ JUPYTER NOTEBOOK ]0.0125 #1029Lordog/dive-into-llms[ JUPYTER NOTEBOOK ]0.00919 #1065Dashboard-Design/Power-BI-Design-Files[ JUPYTER NOTEBOOK ]0.00886 #1263lyogavin/airllm[ JUPYTER NOTEBOOK ]0.00734

METRICS IN CONTEXT

MEDIAN ACROSS ALL 5024 FINDINGS · Δ vs MEDIAN · PERCENTILE = SHARE RANKED BELOW

METRICVALUEMEDIANΔ MEDPERCENTILE

SCORE0.000.00-0.00ABOVE 29%

VELOCITY3.336.67-3.33ABOVE 15%

RETENTION50.0%46.5%+3.5 PPABOVE 51%

FORKS178116+62ABOVE 60%

SURPRISE0.000.01-0.01ABOVE 20%