Language Model Evaluation on MMLU, GSM8K, and BBH
62.07MMLU AccuracyGrad Sim
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Grad Simtraining seed=1337, pipeline=official DATE-LM2025.02 | 62.07 | 56.71 | 64.44 | 61.07 | |
| BipCovembedding backend=BGE emb, training seed=1337, pipeline=official DATE-LM2025.02 | 61.81 | 61.71 | 67.14 | 63.56 | |
| RepSimtraining seed=1337, pipeline=official DATE-LM2025.02 | 61.42 | 58.45 | 66.2 | 62.03 | |
| LESStraining seed=1337, pipeline=official DATE-LM2025.02 | 61.1 | 57.77 | 64.68 | 61.18 | |
| RDS+training seed=1337, pipeline=official DATE-LM2025.02 | 60.63 | 61.41 | 66.23 | 62.75 | |
| Random Avgtraining seed=1337, pipeline=official DATE-LM2025.02 | 60.39 | 59.64 | 66.4 | 62.14 | |
| BM25training seed=1337, pipeline=official DATE-LM2025.02 | 59.85 | 58.98 | 62.63 | 60.49 |