File size: 912 Bytes
09300c3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 |
merge_method: dare_linear # Define the merging method (e.g., linear, dare_linear)
base_model: Undi95/Meta-Llama-3-8B-hf # The base model for the merge
models:
- model: Sao10K/L3-8B-Stheno-v3.2
parameters:
attention_heads: 0.6
weight: 1.0 # Added missing weight parameter
- model: ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.2
parameters:
feedforward: 0.4
weight: 1.0 # Add weight for consistency
- model: O1-OPEN/OpenO1-LLama-8B-v0.1
parameters:
all: 0.3
weight: 1.0 # Add weight for consistency
- model: huihui-ai/Llama-3.1-Tulu-3-8B-abliterated
parameters:
layer_norm: 0.2
weight: 1.0 # Add weight for consistency
parameters:
general: 0.6 # Default parameter for fallback
dtype: float16 # Data type for merging
tokenizer_source: union # Tokenizer combining method
output_path: ./merged_model # Output location for the merged model |