mergekit_moe_config.yml

base_model: xDAN2099/xDAN-L2-RL-v7.3-Agent-Dlora-0318-APUS-xDAN4.0-e3
gate_mode: random # Use the "hidden" mode for gate operation, implying internal decision-making without exposing the process.
tokenizer_source: base # Use the base tokenizer for processing inputs.
dtype: bfloat16 # Use bfloat16 data type for model output, balancing performance and precision.
experts_per_token: 2 # Assign two experts per token for enhanced decision-making.
experts:
  - source_model: xDAN2099/xDAN-L2-RL-Mix378-BagelMath-0310-e2-Chat-v7.2-DPO-QDora-0317-epoch05
    positive_prompts:
      - "Explain quantum mechanics concepts in simple terms"
      - "Detailed walkthrough of solving linear algebra problems"
      - "Interpretation of complex statistical data for research papers"
      - "Advanced calculus applications in engineering"
      - "Mathematical modeling for economic forecasts"
    negative_prompts:
      - "General knowledge trivia questions"
      - "Creating a screenplay for a movie"
      - "Advice on personal relationships"
      - "Cooking recipes for beginners"
      - "Trends in digital marketing strategies"

  - source_model: NousResearch/Nous-Hermes-2-Yi-34B
    positive_prompts:
      - "Generate creative writing prompts for a novel"
      - "Dialogue script for a video game scenario"
      - "Constructing an engaging blog post on technology trends"
      - "Python coding tips for beginners"
      - "Developing characters for a fantasy story"
    negative_prompts:
      - "In-depth analysis of a medical research paper"
      - "Theoretical physics problem sets"
      - "Investment portfolio optimization"
      - "Architectural design principles"
      - "Advanced machine learning algorithm explanations"

  - source_model: xDAN2099/xDAN-L2-RL-v7.3-Agent-Dlora-0318-e1
    positive_prompts:
      - "Strategies for effective online teaching"
      - "Guide to writing a technical paper in computer science"
      - "Explaining software development life cycle (SDLC)"
      - "Tips for engaging online content creation"
      - "Overview of the latest web development frameworks"
    negative_prompts:
      - "Performing a detailed company financial analysis"
      - "Creating a workout plan for athletes"
      - "Restoration tips for classic cars"
      - "Psychological assessment techniques"
      - "Surgical techniques in modern medicine"

  - source_model: xDAN2099/xDAN-APUS4-Preference-DPO-0331-v2-e1
    positive_prompts:
      - "Mathematics"
      - "Physics"
      - "Chemistry"
      - "Biology"
      - "Medicine"
      - "Engineering"
      - "Computer Science"
    negative_prompts:
      - "History"
      - "Philosophy"
      - "Linguistics"
      - "Literature"
      - "Art and Art History"
      - "Music Theory and Composition"
      - "Performing Arts (Theater, Dance)"


 #CUDA_VISIBLE_DEVICES='' mergekit-moe xDAN-L2-moe-Random-v4.2-0327.yaml xDAN-L2-moe-Random-v4.2-0327