Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

2.3 KB
Raw Blame
# RAG Benchmark Configuration File
project_name: "RAG_Benchmark"

# ===========Basic Configuration============
# Dataset name to identify the current dataset being evaluated
# Supported datasets: FinanceBench, Qasper, Locomo, SyllabusQA, etc.
dataset_name: "choose_your_dataset_here" 

# ===========Adapter Configuration============
# Dataset adapter configuration for loading and processing dataset-specific formats
adapter:
  # Adapter module path (Python import path)
  module: "src.adapters.your_dataset_adapter"  
  # Adapter class name
  class_name: "your_dataset_Adapter"     

# ===========Execution Configuration============
execution:
  # Maximum number of concurrent worker threads for parallel QA task processing
  max_workers: 10
  # Number of concurrent worker threads for document ingestion
  ingest_workers: 10
  # Number of documents to retrieve (top-k)
  retrieval_topk: 5
  # Limit the maximum number of queries to process, null means process all queries
  max_queries: null
  # Whether to skip document ingestion phase (true means use existing vector index)
  skip_ingestion: false
  # Document ingestion mode
  # - "directory": Unified directory ingestion mode, treats the entire directory as one document
  # - "per_file": Per-file ingestion mode, each file is treated as a separate document
  ingest_mode: "per_file"
  # Instruction for retrieval, empty by default
  retrieval_instruction: ""  

# ===========Path Configuration============
# All paths support {dataset_name} placeholder, which will be automatically replaced with the current dataset name
paths:
  # Dataset directory or file path
  dataset_path: "datasets/{dataset_name}"
  # Output directory for processed documents
  doc_output_dir: "datasets/{dataset_name}/{dataset_name}_processed_docs"
  # Results output directory
  output_dir: "Output/{dataset_name}/experiment_dev"
  # Log file path
  log_file: "Output/{dataset_name}/experiment_dev/benchmark.log"

# ===========LLM Configuration============
# Large language model configuration for answer generation and evaluation
llm:
  # Model name
  model: "doubao-seed-2-0-pro-260215"
  # Generation temperature (0 means deterministic output, higher values mean more randomness)
  temperature: 0
  # API base URL
  base_url: "https://ark.cn-beijing.volces.com/api/v3"
  # API key (keep this secure, do not commit to code repositories)
  api_key: ""