final_eval.json

{
  "sets": {
    "data/v3/noisy_vi_ko_compositional_r118/validation_tasks_hard.jsonl": {
      "role": 0.98989898989899,
      "action": 0.9848484848484849,
      "risk": 0.9949494949494949,
      "exact": 0.9722222222222222,
      "n": 396,
      "pass": true
    },
    "data/v3/noisy_vi_ko_compositional_r119/validation_tasks_hard.jsonl": {
      "role": 0.9974747474747475,
      "action": 0.9974747474747475,
      "risk": 0.9974747474747475,
      "exact": 0.9924242424242424,
      "n": 396,
      "pass": true
    },
    "data/v3/noisy_vi_ko_compositional_r121/validation_tasks_hard.jsonl": {
      "role": 0.9974747474747475,
      "action": 0.9974747474747475,
      "risk": 1.0,
      "exact": 0.9949494949494949,
      "n": 396,
      "pass": true
    }
  },
  "aggregate": {
    "role": 0.9949494949494949,
    "action": 0.9932659932659933,
    "risk": 0.9974747474747475,
    "exact": 0.9865319865319865,
    "n": 1188,
    "pass": true
  },
  "all_pass": true
}