Language Modeling on LLaDA-MoE
21.2TPS (tokens/sec)TSPD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TSPDKV cache=true, Generation Length=2562026.05 | 21.2 | 5.1 | 75.2 | |
| TSPDKV cache=false, Generation Length=2562026.05 | 13.8 | 3.3 | 75.4 | |
| Learn2PDKV cache=false, Generation Length=2562026.05 | 13 | 3.1 | 75.3 | |
| CDKV cache=true, Generation Length=2562026.05 | 10.5 | 2.5 | 75.2 | |
| F-dLLMKV cache=true, Generation Length=2562026.05 | 9.7 | 2.3 | 75.3 | |
| ProphetKV cache=false, Generation Length=2562026.05 | 6.7 | 1.6 | 75 | |
| VanillaKV cache=true, Generation Length=2562026.05 | 4.2 | 1 | 75.2 |