Skip to content

CONFIG.md

Latest commit

 

History

History
119 lines (103 loc) · 5.1 KB

CONFIG.md

File metadata and controls

119 lines (103 loc) · 5.1 KB
Raw

Configuration (config.yaml)

The behavior of AI Book Weaver is primarily controlled by the config.yaml file. Below is an overview of its main sections and common options.

  • api_settings: Configures the LLM provider and caching.

    • provider: Set to "llama_cpp".
    • llama_cpp:
      • base_url: E.g., "http://localhost:8080".
      • endpoint: (Optional) API endpoint path. Defaults to /v1/chat/completions (recommended for chat and instruction-tuned models with chat templates). Can be set to /completion for raw completion.
      • model: E.g., "llama-3".
      • temperature: Controls randomness (e.g., 0.8).
      • tokenizer_model: (Optional) For client-side token counting if server /tokenize is not used, e.g., "NousResearch/Meta-Llama-3-8B-Instruct".
      • context_window_size: (Optional) Context window size (n_ctx) for llama.cpp.
      • disable_reasoning: (Optional, boolean, default: false) When false, reasoning is enabled and processed by the backend (streaming thinking tokens and logging reasoning), while reasoning text (<think> blocks and reasoning deltas) is automatically stripped from the final generated response. When true, reasoning is explicitly suppressed.
      • llm_options: (Optional) Additional parameters passed directly to llama.cpp (e.g. n_predict, top_k, top_p, repeat_penalty, reasoning_budget).
      • max_retries, retry_delay_seconds: For API call resilience.
      • request_timeout_seconds: Timeout for API requests in seconds.
    • base_cache_dir: Path for API call caching (default: "output/api_cache").
    • base_scratchpad_dir: Path for scratchpad storing characters, locations, and two-pass drafts (default: "output/scratchpad").
    • default_max_retries, default_retry_delay_seconds: Global retry defaults.

  • generation_params: Controls the content and structure of the book.

    • main_topic: (Optional) If empty, a random topic is generated.
    • random_topic_seed: (Optional) Seed for random topic generation.
    • setting: (Optional) Book's setting. Auto-generated if empty.
    • author_name: (Optional) Auto-generated if empty.
    • author_gender: (Optional) male, female, other. Determined or randomized if empty.
    • writing_tone: (Optional) E.g., "academic, informative, yet engaging". Auto-generated if empty.
    • book_title: (Optional) Auto-generated if empty.
    • book_subtitle: (Optional) Auto-generated if empty.
    • length_modifier: (Optional) Adverb like "very", "moderately" to influence summary/outline length.
    • num_chapters_fallback: Default number of chapters if outline generation fails.
    • generate_character_list: (Optional, boolean) true to force, false to disable. If unset, generated for fiction.
    • generate_foreword: (Optional, boolean, default: true).
    • generate_acknowledgements: (Optional, boolean, default: true).
    • generate_appendix: (Optional, boolean, default: true).
    • include_about_author_in_main_book: (Optional, boolean, default: true).

  • style_params: Defines the appearance of the output DOCX.

    • font_name: Default "Times New Roman".
    • font_size: Default 12 (points).
    • page_size_preset: "6x9" (inches) or "A4".
    • margins_mm: top, bottom, left (outside), right (inside), gutter.
    • inline_math_height_multiplier: For scaling inline math images (default: 1.05).
    • display_math_height_inches: Default height for display math images (default: 0.375).
    • inline_math_vertical_offset_half_points: Fine-tunes vertical alignment of inline math (default: -4).

  • output_dir: Directory for the generated book files (default: "output").

  • debug_options:

    • verbose_debug: true or false. Enables detailed LLM call logging and streams responses to the console.
    • save_intermediate_html: true to save the HTML output from Markdown conversion (useful for debugging formatting issues).
    • log_output_path: (Optional) Path for saving file logs (e.g., "output/logs/book.log").

Example config.yaml Snippet:

  • api_settings:
  provider: llama_cpp
  llama_cpp:
    base_url: http://localhost:8080
    model: llama-3
    temperature: 0.8
    context_window_size: 4096
  base_cache_dir: output/api_cache   # Cache stored here
  base_scratchpad_dir: output/scratchpad   # Scratchpad stored here

generation_params:
  # main_topic: "The Secret Life of Urban Foxes" # Leave empty to auto-generate
  # author_name: "Willow Whisperwind" # Leave empty to auto-generate
  length_modifier: concise
  generate_character_list: true # Example for a fictional book

style_params:
  font_name: Garamond
  font_size: 11.5
  page_size_preset: 6x9
  margins_mm:
    top: 19
    bottom: 25
    left: 13  # Outside margin
    right: 19 # Inside margin (for binding)
    gutter: 6 # Space for binding

output_dir: my_books   # Output files will go here

debug_options:
  verbose_debug: true
  save_intermediate_html: true
  log_output_path: output/logs/book.log