config.yaml

architecture:
    llm_model: Viet-Mistral/Vistral-7B-Chat
    embedding_model: intfloat/multilingual-e5-small
    hf_token: NO_USE
    llm_quantized: False
dataset:
    chunk_marker: BK_CHUNK
    required_exts: .docx
    data_dir: ./data
    processed_data_dir: ./processed_data
    signal_type: html_table
    keep_bold: True
retrieve:
    top_k: 2
generation:
    max_new_tokens: 512
    temperature: 0.6
    do_sample: True
    top_p: 0.9
environment:
    device: 'cpu'