dtype: bfloat16 merge_method: passthrough # Layer-Interleaving of Yi-9B-200k # where n = 48 (The number of layers the model originally had) # m = 12 (The number of layers to interleave per row) # M = 36 = 3m (The number of layers interleaved into the model) # s = 48 + 36 = n + M (The number of layers the model will have after the layer interleaving) # P = 15.75B parameters = (s/n)*9 (The number of parameters the model will have after the layer interleaving) slices: - sources: - layer_range: [0, 12] model: /Users/jsarnecki/opt/workspace/01-ai/Yi-9B-200K # Interleaved Layer - sources: - layer_range: [6, 18] model: /Users/jsarnecki/opt/workspace/01-ai/Yi-9B-200K - sources: - layer_range: [12, 24] model: /Users/jsarnecki/opt/workspace/01-ai/Yi-9B-200K # Interleaved Layer - sources: - layer_range: [18, 30] model: /Users/jsarnecki/opt/workspace/01-ai/Yi-9B-200K - sources: - layer_range: [24, 36] model: /Users/jsarnecki/opt/workspace/01-ai/Yi-9B-200K # Interleaved Layer - sources: - layer_range: [30, 42] model: /Users/jsarnecki/opt/workspace/01-ai/Yi-9B-200K - sources: - layer_range: [36, 48] model: /Users/jsarnecki/opt/workspace/01-ai/Yi-9B-200K