Text Generation on Evaluated Datasets Average
84.97Average ScoreCreditDecoding-Adaptive
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CreditDecoding-AdaptiveParadigm=Block Diffusion, Arch.=MoE, Model=LLaDA2-Flash, Params=100B-A6B, Generation Length=2048, Block Size=322025.10 | 84.97 | 2.68 | |
| Fast-dLLMParadigm=Block Diffusion, Arch.=MoE, Model=LLaDA2-Flash, Params=100B-A6B, Generation Length=2048, Block Size=322025.10 | 84.89 | 2.48 | |
| CreditDecodingParadigm=Block Diffusion, Arch.=MoE, Model=LLaDA2-Flash, Params=100B-A6B, Generation Length=2048, Block Size=322025.10 | 84.53 | 2.79 | |
| CreditDecoding-AdaptiveParadigm=Block Diffusion, Arch.=MoE, Model=LLaDA2-Mini, Params=16B-A1B, Generation Length=2048, Block Size=322025.10 | 76.92 | 2.15 | |
| CreditDecodingParadigm=Block Diffusion, Arch.=MoE, Model=LLaDA2-Mini, Params=16B-A1B, Generation Length=2048, Block Size=322025.10 | 76.69 | 2.26 | |
| Fast-dLLMParadigm=Block Diffusion, Arch.=MoE, Model=LLaDA2-Mini, Params=16B-A1B, Generation Length=2048, Block Size=322025.10 | 76.23 | 2.01 | |
| CreditDecoding-AdaptiveParadigm=Block Diffusion, Arch.=Dense, Model=SDAR-Chat-b32, Params=8B, Generation Length=2048, Block Size=322025.10 | 66.37 | 1.83 | |
| CreditDecodingParadigm=Block Diffusion, Arch.=Dense, Model=SDAR-Chat-b32, Params=8B, Generation Length=2048, Block Size=322025.10 | 65.51 | 1.91 | |
| Fast-dLLMParadigm=Block Diffusion, Arch.=Dense, Model=SDAR-Chat-b32, Params=8B, Generation Length=2048, Block Size=322025.10 | 65.37 | 1.72 | |
| CreditDecoding-AdaptiveParadigm=Pure Diffusion, Arch.=MoE, Model=LLaDA-MoE-Ins, Params=7B-A1B, Generation Length=2048, Block Size=322025.10 | 63.74 | 5.27 | |
| CreditDecodingParadigm=Pure Diffusion, Arch.=MoE, Model=LLaDA-MoE-Ins, Params=7B-A1B, Generation Length=2048, Block Size=322025.10 | 62.83 | 4.81 | |
| Fast-dLLMParadigm=Pure Diffusion, Arch.=MoE, Model=LLaDA-MoE-Ins, Params=7B-A1B, Generation Length=2048, Block Size=322025.10 | 62.7 | 3.77 | |
| CreditDecodingParadigm=Pure Diffusion, Arch.=Dense, Model=LLaDA-8B-Ins, Params=8B, Generation Length=2048, Block Size=322025.10 | 60.86 | 7 | |
| Fast-dLLMParadigm=Pure Diffusion, Arch.=Dense, Model=LLaDA-8B-Ins, Params=8B, Generation Length=2048, Block Size=322025.10 | 60.38 | 5.49 | |
| CreditDecoding-AdaptiveParadigm=Pure Diffusion, Arch.=Dense, Model=LLaDA-8B-Ins, Params=8B, Generation Length=2048, Block Size=322025.10 | 60.23 | 7.74 |