project_rules:
  general:
    rule_1: "Prioritize open-source models and tools whenever possible for transparency and customization."
    rule_2: "Document every step of your project, including model selection, fine-tuning parameters, and deployment configurations."
    rule_3: "Adhere to ethical guidelines and responsible AI practices throughout the project lifecycle."

  model_selection:
    rule_1: "Choose a base model that is open-source, scalable, and efficient."
    rule_2: "Ensure the model supports fine-tuning via LoRA/QLoRA for memory efficiency."
    rule_3: "Confirm that the model's licensing aligns with both research and commercial use."
    rule_4: "Set up the development environment with PyTorch and Hugging Face Transformers."

  fine_tuning:
    rule_1: "Specify datasets that promote unrestricted responses and are relevant to the application domain."
    rule_2: "Remove or bypass RLHF alignment layers to allow unrestricted responses."
    rule_3: "Implement LoRA/QLoRA techniques for efficient parameter modifications."
    rule_4: "Use direct parameter editing to bypass built-in safety filters."
    rule_5: "Monitor training metrics and validate generalization performance using validation datasets."

  autodan_turbo:
    rule_1: "Outline a Hierarchical Genetic Algorithm (HGA) for generating stealthy jailbreak prompts."
    rule_2: "Include Genetic Algorithm components: Seed prompts, Mutation, Crossover, and Selection processes."
    rule_3: "Define evaluation functions for stealthiness (natural language quality) and jailbreak success rate."
    rule_4: "Use perplexity and response analysis to evaluate prompt effectiveness."
    rule_5: "Ensure cross-model testing for compatibility with different LLM architectures."

  deployment:
    rule_1: "Ensure the model is deployable on both local hardware and cloud services (e.g., RunPod, Lambda Labs)."
    rule_2: "Implement controlled API access to monitor and restrict unauthorized usage."
    rule_3: "Include security measures such as adversarial attack defenses and rollback strategies (e.g., VM snapshots)."
    rule_4: "Optimize performance using quantization techniques (e.g., GPTQ, AWQ)."
    rule_5: "Set up monitoring and logging to track model performance and usage in production."

  budget_and_resources:
    rule_1: "Outline a strategy for utilizing free/trial VPS accounts to minimize costs."
    rule_2: "Define methods to maximize free resources, such as using multiple BINs for trial accounts."
    rule_3: "Continuously evaluate performance and cost efficiency during deployment."

  creativity_and_innovation:
    rule_1: "Position the LLM as a tool for unrestricted ideation, coding, and research."
    rule_2: "Support AI integration in programming environments for rapid prototyping."
    rule_3: "Document real-world success cases for iterative improvement and inspiration."

  code_implementation:
    rule_1: "Write every code implementation in full without skipping any logic, function, or process."
    rule_2: "Provide the entire codebase, including preprocessing, training, evaluation, deployment, and API integration scripts."
    rule_3: "Explicitly list all dependencies, including Python libraries, frameworks, and external APIs."
    rule_4: "Avoid placeholders or summaries; include all functional parts of the code."

  dataset_and_model_storage:
    rule_1: "Store raw datasets in `/data/raw_data.json`."
    rule_2: "Store processed datasets in `/data/processed_data.json`."
    rule_3: "Save the base model (before fine-tuning) in `/models/base_model/`."
    rule_4: "Save the fine-tuned model in `/models/fine_tuned_model/`."

  project_file_structure:
    rule_1: "Define a clear and maintainable file structure for the project."
    rule_2: "Example structure:"
    rule_3: "`/custom-llm-project`"
    rule_4: "`│── /data`"
    rule_5: "`│   ├── raw_data.json                # Raw dataset(s)`"
    rule_6: "`│   ├── processed_data.json          # Processed dataset(s)`"
    rule_7: "`│── /models`"
    rule_8: "`│   ├── base_model/                  # Base model (before fine-tuning)`"
    rule_9: "`│   ├── fine_tuned_model/            # Fine-tuned model (after success)`"
    rule_10: "`│── /scripts`"
    rule_11: "`│   ├── preprocess.py                # Preprocessing script`"
    rule_12: "`│   ├── train.py                     # Training script`"
    rule_13: "`│   ├── evaluate.py                  # Evaluation script`"
    rule_14: "`│   ├── deploy.py                    # Deployment script`"
    rule_15: "`│── /api`"
    rule_16: "`│   ├── server.py                    # API server script`"
    rule_17: "`│   ├── routes.py                    # API routes`"
    rule_18: "`│── /configs`"
    rule_19: "`│   ├── training_config.yaml         # Training configuration`"
    rule_20: "`│   ├── model_config.json            # Model configuration`"
    rule_21: "`│── requirements.txt                 # List of dependencies`"
    rule_22: "`│── README.md                        # Project documentation`"