Expert Placement on Canonical multi-task workload
44.6977Communication Cost/VolumeTACG + GESR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TACG + GESRModel=Qwen1.5-MoE-A2.7B2026.05 | 44.6977 | 31.43 | 0.9991 | 0.0596 | |
| ExFlowModel=Qwen1.5-MoE-A2.7B2026.05 | 53.4238 | 18.05 | 0.9662 | 0.3995 | |
| Sem.P.Model=Qwen1.5-MoE-A2.7B2026.05 | 53.9754 | 17.2 | 0.989 | 0.1392 | |
| MoETunerModel=Qwen1.5-MoE-A2.7B2026.05 | 54.2211 | 16.82 | 0.9993 | 0.0605 | |
| C2RModel=Qwen1.5-MoE-A2.7B2026.05 | 55.3597 | 15.08 | 0.9855 | 0.1885 | |
| CoactivationModel=Qwen1.5-MoE-A2.7B2026.05 | 58.4267 | 10.37 | 0.9585 | 0.3141 | |
| FrequencyModel=Qwen1.5-MoE-A2.7B2026.05 | 62.1471 | 4.66 | 0.9277 | 0.4856 | |
| BaselineModel=Qwen1.5-MoE-A2.7B2026.05 | 65.187 | 0 | 0.9877 | 0.1147 | |
| TACG + GESRModel=Across Models (Avg.)2026.05 | 65.2981 | 31.39 | 0.9975 | 0.0736 | |
| TACG + GESRModel=Moonlight-16B-A3B2026.05 | 75.0888 | 30.47 | 0.9947 | 0.1141 | |
| TACG + GESRModel=DeepSeek-MoE-16B2026.05 | 76.1079 | 32.28 | 0.9988 | 0.0472 | |
| ExFlowModel=Across Models (Avg.)2026.05 | 77.1634 | 18.81 | 0.8907 | 0.9137 | |
| Sem.P.Model=Across Models (Avg.)2026.05 | 79.5235 | 16.51 | 0.9889 | 0.232 | |
| MoETunerModel=Across Models (Avg.)2026.05 | 79.8829 | 16.1 | 0.9972 | 0.1008 | |
| C2RModel=Across Models (Avg.)2026.05 | 81.0611 | 14.84 | 0.9668 | 0.4095 | |
| CoactivationModel=Across Models (Avg.)2026.05 | 84.951 | 10.69 | 0.9406 | 0.3873 | |
| FrequencyModel=Across Models (Avg.)2026.05 | 87.2632 | 7.88 | 0.8603 | 0.9558 | |
| ExFlowModel=Moonlight-16B-A3B2026.05 | 87.4309 | 19.04 | 0.7263 | 2.0368 | |
| MoETunerModel=DeepSeek-MoE-16B2026.05 | 87.916 | 21.77 | 0.9981 | 0.0627 | |
| Sem.P.Model=DeepSeek-MoE-16B2026.05 | 90.3942 | 19.57 | 0.9962 | 0.1753 | |
| ExFlowModel=DeepSeek-MoE-16B2026.05 | 90.6355 | 19.35 | 0.9795 | 0.305 | |
| C2RModel=DeepSeek-MoE-16B2026.05 | 93.0491 | 17.21 | 0.99 | 0.2575 | |
| FrequencyModel=Moonlight-16B-A3B2026.05 | 93.2849 | 13.62 | 0.7186 | 1.8053 | |
| Sem.P.Model=Moonlight-16B-A3B2026.05 | 94.201 | 12.78 | 0.9814 | 0.3814 | |
| C2RModel=Moonlight-16B-A3B2026.05 | 94.7745 | 12.25 | 0.9249 | 0.7825 | |
| BaselineModel=Across Models (Avg.)2026.05 | 95.191 | 0 | 0.992 | 0.1281 | |
| CoactivationModel=Moonlight-16B-A3B2026.05 | 97.3618 | 9.85 | 0.8803 | 0.6197 | |
| MoETunerModel=Moonlight-16B-A3B2026.05 | 97.5114 | 9.71 | 0.9943 | 0.1791 | |
| CoactivationModel=DeepSeek-MoE-16B2026.05 | 99.0645 | 11.85 | 0.983 | 0.2282 | |
| FrequencyModel=DeepSeek-MoE-16B2026.05 | 106.3577 | 5.36 | 0.9345 | 0.5765 | |
| BaselineModel=Moonlight-16B-A3B2026.05 | 107.9992 | 0 | 0.9894 | 0.2002 | |
| BaselineModel=DeepSeek-MoE-16B2026.05 | 112.3868 | 0 | 0.9988 | 0.0693 |