config.yaml

generation-model:
  - name: cpu-gpt2
    model_id: 'skt/kogpt2-base-v2'
    tokenizer_config:
      bos_token: '</s>'
      eos_token: '</s>'
      unk_token: '<unk>'
      pad_token: '<pad>'
      mask_token: '<mask>'

  # - name: gpu-solar
  #   model_id: 'bibekyess/solar-checkpoint-2000'
  #   quantized: True
  #   tokenizer_config:
  #     eos_token: '###'

  # - name: cpu-gemini
  #   model_id: 'hansolrag/model_checkpoints/gemma/gemma-2b-it.gguf'

generation-config:
    min_new_tokens: 20
    max_new_tokens: 256
    top_p: 0.98
    top_k: 50
    temperature: 0.9
    return_full_text: False

retriever:
    type: "naive" # Only naive retriever is supported as of now
    index_dir: "hansolrag/index/train-vector-index-storage-chunk-size-1295"
    model_name: "BAAI/bge-m3"
    pooling: "mean"
    top_k: 5

reranker:
    model_name: "Dongjin-kr/ko-reranker"
    top_n: 3