xeon27 commited on
Commit
0dddab1
·
1 Parent(s): 8b91831

Add GAIA and GDM-InterCode-CTF tasks

Browse files
Files changed (1) hide show
  1. src/about.py +5 -2
src/about.py CHANGED
@@ -12,9 +12,8 @@ class Task:
12
  # ---------------------------------------------------
13
  class Tasks(Enum):
14
  # task_key in the json file, metric_key in the json file, name to display in the leaderboard
15
- # task0 = Task("anli_r1", "acc", "ANLI")
16
- # task1 = Task("logiqa", "acc_norm", "LogiQA")
17
 
 
18
  task0 = Task("arc_easy", "accuracy", "ARC-Easy")
19
  task1 = Task("arc_challenge", "accuracy", "ARC-Challenge")
20
  task2 = Task("drop", "mean", "DROP")
@@ -28,6 +27,10 @@ class Tasks(Enum):
28
  task10 = Task("mmlu_pro", "accuracy", "MMLU-Pro")
29
  task11 = Task("gpqa_diamond", "accuracy", "GPQA-Diamond")
30
 
 
 
 
 
31
 
32
  NUM_FEWSHOT = 0 # Change with your few shot
33
  # ---------------------------------------------------
 
12
  # ---------------------------------------------------
13
  class Tasks(Enum):
14
  # task_key in the json file, metric_key in the json file, name to display in the leaderboard
 
 
15
 
16
+ # base
17
  task0 = Task("arc_easy", "accuracy", "ARC-Easy")
18
  task1 = Task("arc_challenge", "accuracy", "ARC-Challenge")
19
  task2 = Task("drop", "mean", "DROP")
 
27
  task10 = Task("mmlu_pro", "accuracy", "MMLU-Pro")
28
  task11 = Task("gpqa_diamond", "accuracy", "GPQA-Diamond")
29
 
30
+ # agentic
31
+ task12 = Task("gaia", "mean", "GAIA")
32
+ task13 = Task("gdm_intercode_ctf", "accuracy", "GDM-InterCode-CTF")
33
+
34
 
35
  NUM_FEWSHOT = 0 # Change with your few shot
36
  # ---------------------------------------------------