name: rag
version: "paper-1"
adapter: rag
sample_workers: 2
chunking:
  strategy: token
  chunk_size: 256
  chunk_overlap: 32
  tokenizer_model: gpt-4
  expand_turns_on_display: true
  dedup_turns_on_display: false
embedding:
  provider: openai
  model: text-embedding-3-small
  api_key: "${VECTORIZE_API_KEY}"
  base_url: "${VECTORIZE_BASE_URL:https://api.openai.com/v1}"
  batch_size: 128
  max_retries: 3
storage:
  backend: local
  save_every_conversations: 10
rerank:
  enabled: true
  model: rerank-v3.5
  api_key: "${RERANK_API_KEY}"
  base_url: "${RERANK_BASE_URL:https://api.cohere.com}"
  endpoint: "${RERANK_ENDPOINT:/v2/rerank}"
  timeout_seconds: 120
  max_retries: 3
llm:
  provider: openai
  model: gpt-4o-mini
  api_key: "${LLM_API_KEY}"
  base_url: "${LLM_BASE_URL:https://api.openai.com/v1}"
  temperature: 0
  max_tokens: 16384
search:
  embedding_top_k: 128
  rerank_top_k: 15
  sort_selected_chunks_by_time: false
  dynamic_top_k:
    enabled: false
    score_threshold: 0.5
    fallback_top_k: 3
  num_workers: 8
answer:
  prompt_key: answer_prompt_cot
  num_workers: 50
  max_retries: 3
dataset_overrides:
  meme:
    search:
      sort_selected_chunks_by_time: true
    answer:
      prompt_key: answer_prompt_meme_official
  memoryagentbench:
    chunking:
      expand_turns_on_display: false
    sources:
      factconsolidation_sh_262k:
        chunking:
          strategy: turn
          chunk_size: 1
          chunk_overlap: 0
      factconsolidation_mh_262k:
        chunking:
          strategy: turn
          chunk_size: 1
          chunk_overlap: 0