Overall Performance on Aggregated All Benchmarks
40.3Average ScoreDiReCT
Evaluation Results
| Method | Links | |
|---|---|---|
| DiReCTBackbone=Llama-1.1B2026.05 | 40.3 | |
| InfoBatchBackbone=Llama-1.1B2026.05 | 40.1 | |
| GradNorm (IS)Backbone=Llama-1.1B2026.05 | 38.9 | |
| Loss-basedBackbone=Llama-1.1B2026.05 | 38.3 | |
| Perplexity-basedBackbone=Llama-1.1B2026.05 | 38.2 | |
| Uniform SamplingBackbone=Llama-1.1B2026.05 | 37.9 | |
| DiReCTBackbone=GPT-2-Medium (355M)2026.05 | 32.3 | |
| InfoBatchBackbone=GPT-2-Medium (355M)2026.05 | 32.1 | |
| GradNorm (IS)Backbone=GPT-2-Medium (355M)2026.05 | 31.3 | |
| Perplexity-basedBackbone=GPT-2-Medium (355M)2026.05 | 31 | |
| Loss-basedBackbone=GPT-2-Medium (355M)2026.05 | 30.4 | |
| Uniform SamplingBackbone=GPT-2-Medium (355M)2026.05 | 30.2 |