brunneis commited on
Commit
a2214d4
·
unverified ·
1 Parent(s): 53ae66b

Update score definition

Browse files
Files changed (1) hide show
  1. src/leaderboard/read_evals.py +2 -2
src/leaderboard/read_evals.py CHANGED
@@ -116,8 +116,8 @@ class EvalResult:
116
  """Converts the Eval Result to a dict compatible with our dataframe display"""
117
  # average = sum([v for v in self.results.values() if v is not None]) / len(Tasks)
118
  solbench = (
119
- self.results.get('naive_judge', 0) * 0.2 +
120
- self.results.get('human_eval_solidity', 0) * 0.9
121
  )
122
  data_dict = {
123
  "eval_name": self.eval_name, # not a column, just a save name,
 
116
  """Converts the Eval Result to a dict compatible with our dataframe display"""
117
  # average = sum([v for v in self.results.values() if v is not None]) / len(Tasks)
118
  solbench = (
119
+ self.results.get('naive_judge', 0) * 0.3 +
120
+ self.results.get('human_eval_solidity', 0) * 0.7
121
  )
122
  data_dict = {
123
  "eval_name": self.eval_name, # not a column, just a save name,