params.yaml

log_level: INFO
n_bootstrap: 1_000
plot_format: png

weighting:
  - weight_col: equal_task_weight
    graph_snippet: Equally weighted tasks
  - weight_col: invsqrt_task_weight
    graph_snippet: Tasks weighted by 1/sqrt(family size)
  - weight_col: null
    graph_snippet: "None"

stages:
  filter_aird_runs:
    task_families:
      - ai_rd_fix_embedding
      - ai_rd_nanogpt_chat_rl
      - ai_rd_optimize_llm_foundry
      - ai_rd_restricted_mlm
      - ai_rd_rust_codecontests_inference
      - ai_rd_small_scaling_law
      - ai_rd_triton_cumsum

  wrangle_bar_by_time_allocation:
    time_limits: [1800, 7200, 28800]

plots:
  colors:
    agent_aliases:
      "Claude 3 Opus":
        light: "#fe55b8"
        base: "#c90076"
        dark: "#8f0355"
      "Claude 3.5 Sonnet (Old)":
        light: "#df4166"
        base: "#B61F43"
        dark: "#80162F"
      "Claude 3.5 Sonnet (New)":
        light: "#da8f69"
        base: "#CE6A38"
        dark: "#924924"
      "o1":
        light: "#29F4a9"
        base: "#05AC6F"
        dark: "#04794E"
      "o1-preview":
        light: "#138Bdb"
        base: "#0C5789"
        dark: "#083C5F"
      "GPT-4 0314":
        light: "#9B6B9E"
        base: "#744474"
        dark: "#4D2D4D"
      "GPT-4 Turbo":
        light: "#45B3D6"
        base: "#2B8FB0"
        dark: "#1A5668"
      "GPT-4o":
        light: "#98D47E"
        base: "#6AAD52"
        dark: "#456F35"
      human:
        light: "#c9c9c9"
        base: "#858585"
        dark: "#484848"
    default: "#000000"

  legend_order:
    - Claude 3.5 Sonnet (Old)
    - Claude 3.5 Sonnet (New)
    - Claude 3 Opus
    - o1-preview
    - o1
    - GPT-4 0314
    - GPT-4 Turbo
    - GPT-4o
    - Human 8-hour score