Blame for eval/package_check.json

{
 "device": "mps",
 "bench": {
  "cases": 300,
  "questions": 1200,
  "agreement": 1.0,
  "max_abs_prob_diff": 0.0
 },
 "td": {
  "cases": 300,
  "questions": 1500,
  "agreement": 1.0,
  "max_abs_prob_diff": 0.0001
 },
 "nd_amazon": {
  "cases": 300,
  "questions": 300,
  "agreement": 1.0,
  "max_abs_prob_diff": 0.0
 }
}