Instruction Tuning on BBH
66.2Accuracy (BBH)Full Data
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Full DataModel=Qwen-2.5-7B-Instruct, Selection Ratio=100%, Cost (h)=32.82026.05 | 66.2 | 70.1 | — | |
| PODSModel=Qwen-2.5-7B-Instruct, Selection Ratio=50%, Cost (h)=16.9, Speedup=1.9×2026.05 | 66.2 | 70.1 | 0 | |
| TACSPool=COT, Backbone=Llama-3.2-3B2026.05 | 49.44 | — | — | |
| TACSPool=Oasst, Backbone=Llama-3.2-3B2026.05 | 49.38 | — | — | |
| TACSPool=Average, Backbone=Llama-3.2-3B2026.05 | 48.95 | — | — | |
| TACSPool=Flan, Backbone=Llama-3.2-3B2026.05 | 48.92 | — | — | |
| LESSPool=COT, Backbone=Llama-3.2-3B2026.05 | 48.4 | — | — | |
| Full DataModel=LLaMA-3.2-3B, Selection Ratio=100%, Cost (h)=17.62026.05 | 48.4 | 52.2 | — | |
| PODSModel=LLaMA-3.2-3B, Selection Ratio=50%, Cost (h)=9.0, Speedup=2.0×2026.05 | 48.4 | 52.2 | 0 | |
| LESSPool=Dolly, Backbone=Llama-3.2-3B2026.05 | 48.24 | — | — | |
| TACSPool=Dolly, Backbone=Llama-3.2-3B2026.05 | 48.06 | — | — | |
| ToVPool=Oasst, Backbone=Llama-3.2-3B2026.05 | 48.06 | — | — | |
| ToVPool=Flan, Backbone=Llama-3.2-3B2026.05 | 47.62 | — | — | |
| ToVPool=Average, Backbone=Llama-3.2-3B2026.05 | 47.55 | — | — | |
| LESSPool=Average, Backbone=Llama-3.2-3B2026.05 | 47.49 | — | — | |
| ToVPool=Dolly, Backbone=Llama-3.2-3B2026.05 | 47.28 | — | — | |
| ToVPool=COT, Backbone=Llama-3.2-3B2026.05 | 47.25 | — | — | |
| LESSPool=Flan, Backbone=Llama-3.2-3B2026.05 | 47.04 | — | — | |
| LESSPool=Oasst, Backbone=Llama-3.2-3B2026.05 | 46.3 | — | — | |
| RandomPool=Oasst, Backbone=Llama-3.2-3B2026.05 | 46.11 | — | — | |
| RandomPool=Dolly, Backbone=Llama-3.2-3B2026.05 | 46.02 | — | — | |
| RandomPool=Flan, Backbone=Llama-3.2-3B2026.05 | 45.83 | — | — | |
| RandomPool=Average, Backbone=Llama-3.2-3B2026.05 | 45.76 | — | — | |
| RandomPool=COT, Backbone=Llama-3.2-3B2026.05 | 45.09 | — | — |