mirror of
https://github.com/Nighthawk42/MiraiAssist.git
synced 2026-08-30 08:32:26 +00:00
104 lines
4.7 KiB
YAML
104 lines
4.7 KiB
YAML
# ================================================
|
|
# MireiAssist Configuration File
|
|
# ================================================
|
|
|
|
# --- LLM (Large Language Model) Settings ---
|
|
llm:
|
|
api_base_url: "http://localhost:1234/v1" # e.g., LM Studio, Ollama, OpenAI
|
|
api_key_env_var: "NONE" # Environment variable for API key. "NONE" or null if not needed.
|
|
model_name: "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF" # Model identifier for the API
|
|
# !!! SET THIS ACCURATELY FOR YOUR MODEL !!! (e.g., Llama3 8k=8192, GPT-4 128k=131072)
|
|
# Used for Tiktoken length checks if available. Set to 0 to disable checks.
|
|
model_context_window: 8192
|
|
system_prompt: "You are Mirei, a helpful and concise AI assistant integrated into a desktop application. Respond clearly and directly. You can use markdown formatting."
|
|
temperature: 0.7 # 0=deterministic, >1 more creative
|
|
max_tokens: 1536 # Max tokens for the LLM's *response* (ensure less than model_context_window)
|
|
timeout_seconds: 120.0
|
|
max_retries: 1
|
|
|
|
# --- Context Manager (RAG) Settings ---
|
|
context_manager:
|
|
storage_path: "data/conversation_state.json" # Stores full raw history
|
|
vector_db_path: "data/chroma_db" # Stores vector index
|
|
# Embedding Model: https://www.sbert.net/docs/pretrained_models.html
|
|
embedding_model_name: "all-MiniLM-L6-v2" # Model for text embeddings ("multi-qa-MiniLM-L6-cos-v1" is another option)
|
|
collection_name: "mirei_chat_history" # ChromaDB collection name
|
|
retrieval_results: 3 # How many relevant history chunks to retrieve
|
|
include_recent_messages: 2 # How many recent messages to *always* include with RAG results
|
|
|
|
# --- STT (Speech-to-Text) Settings ---
|
|
stt:
|
|
# Faster Whisper: https://github.com/SYSTRAN/faster-whisper#model-weights
|
|
model_size: "base.en" # tiny.en, base.en, small.en, medium.en, large-v3
|
|
device: "cpu" # cpu, cuda (Ensure PyTorch with CUDA is installed if using cuda)
|
|
compute_type: "int8" # CPU: int8, int16, float32. CUDA: float16, int8_float16, int8
|
|
vad_filter: true # Enable Voice Activity Detection (recommended)
|
|
vad_parameters: # Optional VAD tuning (see faster-whisper docs)
|
|
# threshold: 0.5
|
|
# min_silence_duration_ms: 50
|
|
# min_speech_duration_ms: 250
|
|
# speech_pad_ms: 400
|
|
beam_size: 5 # Higher = more accurate but slower
|
|
language: null # null = auto-detect, or "en", "ja", etc.
|
|
initial_prompt: null # Optional prompt to guide transcription
|
|
|
|
# --- TTS (Text-to-Speech) Settings ---
|
|
tts:
|
|
# Kokoro: https://github.com/hexgrad/Kokoro/tree/main#-voices
|
|
lang_code: "a" # a=American, b=British, j=Japanese
|
|
voice: "af_heart" # Kokoro voice preset
|
|
speed: 1.0 # 1.0 = normal speed
|
|
|
|
# --- Audio Settings ---
|
|
audio:
|
|
sample_rate: 16000
|
|
record_chunk_size: 1024
|
|
channels: 1 # Mono
|
|
input_device: null # null = default device, or index (e.g., 1), or partial name ("Microphone Array")
|
|
output_device: null # null = default device, or index (e.g., 4), or partial name ("Speakers")
|
|
|
|
# --- Activation Settings ---
|
|
activation:
|
|
push_to_talk_key: "<Control-space>" # Tkinter format. "" or null to disable.
|
|
|
|
# --- GUI Settings ---
|
|
gui:
|
|
theme_preference: "dark" # system, light, dark
|
|
|
|
# --- Logging Settings ---
|
|
logging:
|
|
# General Settings
|
|
console_log_level: "INFO" # Minimum level for console (DEBUG, INFO, WARNING, ERROR, CRITICAL)
|
|
|
|
# File Logging Settings
|
|
file_logging_enabled: true
|
|
log_directory: "logs"
|
|
application_log_file: "mirai_assist.log"
|
|
application_log_level: "DEBUG" # Capture detailed info for debugging
|
|
error_log_file: "errors.log"
|
|
error_log_level: "WARNING" # Capture warnings and errors separately
|
|
log_max_bytes: 5242880 # 5 MB
|
|
log_backup_count: 3
|
|
|
|
# Rich Console Logging Settings (Requires 'rich' library: uv add rich)
|
|
rich_console_logging: true # Set to true to use Rich for console output
|
|
rich_show_time: true # Show timestamp in Rich console output
|
|
rich_show_level: true # Show log level (e.g., INFO, ERROR)
|
|
rich_show_path: false # Show module path (e.g., modules.llm_manager) - can be verbose
|
|
rich_markup: true # Enable Rich markup processing (e.g., [bold], [red]) in log messages
|
|
rich_tracebacks: true # Use Rich formatting for exception tracebacks
|
|
rich_tracebacks_show_locals: false # Optionally show local variables in tracebacks
|
|
rich_keywords: # Optional: List of keywords to highlight in log messages (case-sensitive)
|
|
- "CRITICAL"
|
|
- "ERROR"
|
|
# - "WARNING"
|
|
- "Failed"
|
|
- "Success"
|
|
- "Ready"
|
|
- "Initialized"
|
|
- "Shutdown"
|
|
- "Saving"
|
|
- "Loading"
|
|
- "RAG"
|
|
- "Retrieving"
|
|
- "Indexing" |