Mathematical Reasoning on MATH (Pass@1, Pass@100)
46.2Pass@1LaDiR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LaDiRBackbone=LLAMA 3.1 8B2025.10 | 46.2 | 63.7 | |
| TaH+Backbone=LLAMA 3.1 8B2025.10 | 46.1 | 49.4 | |
| Soft TokenBackbone=LLAMA 3.1 8B2025.10 | 44.7 | — | |
| Soft ThinkBackbone=LLAMA 3.1 8B2025.10 | 44.3 | 46.7 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=0.12025.10 | 43.8 | 46.8 | |
| Discrete LatentBackbone=LLAMA 3.1 8B2025.10 | 43.2 | 47.3 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=0.72025.10 | 42.9 | 49.4 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=12025.10 | 42.1 | 51 | |
| SFTBackbone=LLAMA 3.1 8B, Decoding temperature (alpha)=1.22025.10 | 41.2 | 52.5 | |
| CoT SFTBackbone=LLADA 8B2025.10 | 39 | 51 | |
| CODIBackbone=LLAMA 3.1 8B2025.10 | 38.5 | 45.1 | |
| CoconutBackbone=LLAMA 3.1 8B2025.10 | 37.3 | 39.3 | |
| Base ModelBackbone=LLADA 8B2025.10 | 36.2 | 48.6 | |
| iCoTBackbone=LLAMA 3.1 8B2025.10 | 35.2 | 37.9 | |
| LaDiR –w/o Stage 2Backbone=LLAMA 3.1 8B, Training Stage=without Stage 22025.10 | 30.7 | 35.8 | |
| Pause TokenBackbone=LLAMA 3.1 8B2025.10 | 27.2 | 31.4 | |
| Sol-Only SFTBackbone=LLAMA 3.1 8B2025.10 | 13.3 | 18.9 | |
| LD4LGBackbone=LLAMA 3.1 8B2025.10 | 9.1 | 15.3 | |
| PLANNERBackbone=LLAMA 3.1 8B2025.10 | 5.7 | 13.1 |