Multi-Thinker Partitioned Problem Solving on Synthetic Partitioned Dataset (N=2, 100 problems) (test)
94.7AccuracyCanonicalMerge
Evaluation results
| Method | Links | |
|---|---|---|
| CanonicalMergeregime=query-known, token budget=20482026.07 | 94.7 | |
| CanonicalMergearm=A1, regime=q-known, tau=1.0, token budget=5122026.07 | 94 | |
| CanonicalMergearm=A3, regime=q-blind, tau=1.0, token budget=5122026.07 | 85 | |
| PackLLMregime=query-known, token budget=20482026.07 | 50 | |
| PackLLMarm=A1, regime=q-known, tau=1.0, token budget=5122026.07 | 38 | |
| PackLLMarm=A3, regime=q-blind, tau=1.0, token budget=5122026.07 | 34 | |
| PackLLMarm=A4, regime=q-blind, tau=0.01 (one-hot floor), token budget=5122026.07 | 1 |