Textual Reasoning on Averaged
85.63TPSSTDec
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| STDecBase Model=Dream-7B-Instruct, Decoding Strategy=STDec2026.04 | 85.63 | 10.03 | 55.16 | |
| LocalLeapBase Model=Dream-7B-Instruct, Decoding Strategy=LocalLeap2026.04 | 67.93 | 7.95 | 54.44 | |
| STDecBackbone=LLaDA-8B-Instruct2026.04 | 59.09 | 7.6 | 46.87 | |
| LLaDA + LocalLeapBackbone=LLaDA-8B-Instruct2026.04 | 53.67 | 6.91 | 46.17 | |
| Fast-dLLM (Parallel)Base Model=Dream-7B-Instruct, Decoding Strategy=Fast-dLLM (Parallel)2026.04 | 50.03 | 5.86 | 55.79 | |
| LLaDA + Fast-dLLM (Parallel)Backbone=LLaDA-8B-Instruct2026.04 | 42.99 | 5.53 | 46.95 | |
| Half-StepBase Model=Dream-7B-Instruct, Decoding Strategy=Half-Step2026.04 | 17.08 | 2 | 43.42 | |
| LLaDA + Half-StepBackbone=LLaDA-8B-Instruct2026.04 | 15.55 | 2 | 43.25 | |
| dKV-CacheBase Model=Dream-7B-Instruct, Decoding Strategy=dKV-Cache2026.04 | 12.99 | 1.52 | 52.43 | |
| LLaDA + dKV-CacheBackbone=LLaDA-8B-Instruct2026.04 | 12.06 | 1.55 | 47.36 | |
| Vanilla DreamBase Model=Dream-7B-Instruct, Decoding Strategy=Vanilla2026.04 | 8.54 | 1 | 54.3 | |
| Vanilla LLaDABackbone=LLaDA-8B-Instruct2026.04 | 7.77 | 1 | 46.59 |