models:
  - model: deepseek-ai/DeepSeek-R1-Distill-Llama-70B
  - model: unreleased-novatempus-70b-v0.1.1
merge_method: slerp
base_model: deepseek-ai/DeepSeek-R1-Distill-Llama-70B
parameters:
  t:
    - filter: self_attn
      value: [0.2, 0.25, 0.3, 0.25, 0.2]
    - filter: "q_proj|k_proj|v_proj"
      value: [0.2, 0.25, 0.3, 0.25, 0.2]
    - filter: "up_proj|down_proj"
      value: [0.2, 0.3, 0.4, 0.3, 0.2]
    - filter: mlp
      value: [0.25, 0.35, 0.55, 0.35, 0.25]
    - value: 0.45  # default for other components
dtype: bfloat16
tokenizer:
  source: deepseek-ai/DeepSeek-R1-Distill-Llama-70B #necessary to fix tokenizer