dtype: bfloat16 | |
merge_method: passthrough | |
# Layer-Interleaving of AetherResearch/Cerebrum-1.0-7b | |
# where n = 32 (The number of layers the model originally had) | |
# m = 8 (The number of layers to interleave per row) | |
# M = 24 = 3m (The number of layers interleaved into the model) | |
# s = 56 = 32 + 24 = n + M (The number of layers the model will have after the layer interleaving) | |
# P = 12.25B parameters = (s/n)*7 (The number of parameters the model will have after the layer interleaving) | |
slices: | |
- sources: | |
- layer_range: [0, 8] | |
model: /Users/jsarnecki/opt/Workspace/AetherResearch/Cerebrum-1.0-7b | |
# Interleaved Layer | |
- sources: | |
- layer_range: [4, 12] | |
model: /Users/jsarnecki/opt/Workspace/AetherResearch/Cerebrum-1.0-7b | |
- sources: | |
- layer_range: [8, 16] | |
model: /Users/jsarnecki/opt/Workspace/AetherResearch/Cerebrum-1.0-7b | |
# Interleaved Layer | |
- sources: | |
- layer_range: [12, 20] | |
model: /Users/jsarnecki/opt/Workspace/AetherResearch/Cerebrum-1.0-7b | |
- sources: | |
- layer_range: [16, 24] | |
model: /Users/jsarnecki/opt/Workspace/AetherResearch/Cerebrum-1.0-7b | |
# Interleaved Layer | |
- sources: | |
- layer_range: [20, 28] | |
model: /Users/jsarnecki/opt/Workspace/AetherResearch/Cerebrum-1.0-7b | |
- sources: | |
- layer_range: [24, 32] | |
model: /Users/jsarnecki/opt/Workspace/AetherResearch/Cerebrum-1.0-7b | |