base_model: d:/DeepSeek-R1-Distill-Qwen-1.5B-uncensored
gate_mode: random
architecture: qwen
dtype: float32
experts_per_token: 4
experts:
  - source_model: D:/DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1
  - source_model: D:/Qwen2.5-DeepScaleR-1.5B-Preview
  - source_model: D:/QwQ-R1-Distill-1.5B-CoT
  - source_model: D:/DeepSeek-R1-ReDistill-Qwen-1.5B-v1.0
  - source_model: D:/DeepSeek-R1-Distill-Qwen-1.5B-uncensored
  - source_model: D:/Qwen2.5-1.5B-Instruct

shared_experts:
  - source_model: D:/DeepSeek-R1-Distill-Qwen-1.5B-uncensored
    residual_scale: 0.01 # downweight output from shared expert to prevent overcooking the model