| license: apache-2.0 | |
| datasets: | |
| - HuggingFaceTB/cosmopedia | |
| An untrained precursor MoE created from Cosmo using mergekit. | |
| Gate routing initialized using prompt hidden state method. Five are based on the visualized topic clusters of Cosmopedia data, three are task-oriented. | |
| Degenerate layers were 0, 1, and 2. Expert gates for layers 0, 1, and 2 have been randomly initialized to with luck mitigate this. | |