List your company

Multi-step tensor program optimization on Step-TP (100 held-out tensor programs)

103.17Avg SamplesBeam Search
12.461236.010659.5683.1094May 25, 2026

Evaluation results

MethodLinks
Beam SearchModel=Qwen3-32B trained on Step-TP, Hardware=H20-3e2026.05103.1711442.94.57561.820.6
Greedy SearchModel=Qwen3-32B trained on Step-TP, Hardware=H20-3e2026.0535.914120.791.78193.550.58
BFS SearchModel=Qwen3-32B trained on Step-TP, Hardware=H20-3e2026.0528.273118.272.11173.130.65
DFS SearchModel=Qwen3-32B trained on Step-TP, Hardware=H20-3e2026.0528.113120.971.68242.620.75
MCTSModel=Qwen3-32B trained on Step-TP, Hardware=H20-3e2026.0517.412124.964.54171.91.43
Chain SearchModel=Qwen3-32B trained on Step-TP, Hardware=A1002026.0516.852122.013.62265.351.31
Chain SearchModel=Qwen3-32B trained on Step-TP, Hardware=H20-3e2026.0516.632123.622.05286.921.42
Chain SearchModel=Qwen3-32B trained on Step-TP, Hardware=H20-3e, Parent Nodes=Without2026.0515.952120.254.9696.141.63