{
 "runs/m2w31\\m2w31_domain.json": {
  "split": "test_domain",
  "model": "google/gemma-4-31B-it",
  "adapter": "emrevrg/AUBIN-31B",
  "protocol": "MindAct multi-choice (groups of 5 + None, iterative), top-50 by official ranker",
  "summary": {
   "n": 200,
   "element_accuracy": 0.54,
   "operation_f1": 0.8646,
   "step_success_rate": 0.485,
   "pos_in_topk_rate": 0.9,
   "minutes": 96.5
  },
  "train": null
 },
 "runs/m2w31\\m2w31_website.json": {
  "split": "test_website",
  "model": "google/gemma-4-31B-it",
  "adapter": "emrevrg/AUBIN-31B",
  "protocol": "MindAct multi-choice (groups of 5 + None, iterative), top-50 by official ranker",
  "summary": {
   "n": 200,
   "element_accuracy": 0.41,
   "operation_f1": 0.8202,
   "step_success_rate": 0.335,
   "pos_in_topk_rate": 0.905,
   "minutes": 108.7
  },
  "train": null
 },
 "runs/sat/aubin-sat-12ef\\m2wa_test_domain.json": {
  "split": "test_domain",
  "model": "google/gemma-4-12B-it",
  "adapter": "hf:emrevrg/AUBIN-12B",
  "protocol": "MindAct multi-choice (groups of 5 + None, iterative), top-50 by official ranker",
  "summary": {
   "n": 200,
   "element_accuracy": 0.475,
   "operation_f1": 0.8767,
   "step_success_rate": 0.45,
   "pos_in_topk_rate": 0.9,
   "minutes": 228.6
  },
  "train": {
   "steps": 1200,
   "items": 7995,
   "lr": 3e-05,
   "bs": 4,
   "minutes": 185.3,
   "best_dev": [
    0.8133333333333334,
    1000
   ],
   "adapter_dir": "/kaggle/working/lora_m2wa"
  }
 },
 "runs/sat/aubin-sat-12ef\\m2wa_test_task.json": {
  "split": "test_task",
  "model": "google/gemma-4-12B-it",
  "adapter": "/kaggle/working/lora_m2wa",
  "protocol": "MindAct multi-choice (groups of 5 + None, iterative), top-50 by official ranker",
  "summary": {
   "n": 200,
   "element_accuracy": 0.48,
   "operation_f1": 0.8145,
   "step_success_rate": 0.4,
   "pos_in_topk_rate": 0.945,
   "minutes": 35.6
  },
  "train": null
 },
 "runs/sat/aubin-sat-12ef\\m2wa_test_website.json": {
  "split": "test_website",
  "model": "google/gemma-4-12B-it",
  "adapter": "/kaggle/working/lora_m2wa",
  "protocol": "MindAct multi-choice (groups of 5 + None, iterative), top-50 by official ranker",
  "summary": {
   "n": 200,
   "element_accuracy": 0.375,
   "operation_f1": 0.8162,
   "step_success_rate": 0.32,
   "pos_in_topk_rate": 0.905,
   "minutes": 34.6
  },
  "train": null
 }
}