| models: | |
| - model: NousResearch/Meta-Llama-3-8B | |
| # Base model providing a general foundation without specific parameters | |
| - model: NousResearch/Meta-Llama-3-8B-Instruct | |
| parameters: | |
| density: 0.60 # Increased density to emphasize instruction-based features | |
| weight: 0.25 # Reduced weight to balance total sum | |
| - model: winglian/llama-3-8b-256k-PoSE | |
| parameters: | |
| density: 0.60 # Emphasizing extended context handling | |
| weight: 0.20 # Adjusted to balance the total weight sum | |
| - model: Locutusque/Llama-3-Orca-1.0-8B | |
| parameters: | |
| density: 0.55 # Standard density for effective integration | |
| weight: 0.15 # Reduced weight to allocate more to critical models | |
| - model: abacusai/Llama-3-Smaug-8B | |
| parameters: | |
| density: 0.55 # Consistent density for uniform behavior | |
| weight: 0.15 # Reduced weight for balance and efficiency | |
| - model: beomi/Llama-3-Open-Ko-8B-Instruct-preview | |
| parameters: | |
| density: 0.55 # Adjusted density to fit within the ensemble | |
| weight: 0.30 # Increased weight to emphasize this Korean-based model | |
| merge_method: dare_ties | |
| base_model: NousResearch/Meta-Llama-3-8B | |
| parameters: | |
| int8_mask: true | |
| dtype: bfloat16 | |