Hugging Face's logo

Spaces:
hallucinations-leaderboard
/
leaderboard
Running on CPU Upgrade

App Files Community
27
pminervini commited on
Commit
11257ec
•
1 Parent(s): b99996b

update

Browse files
Files changed (3) hide show
  1. src/backend/tasks/halueval/halueval_dialogue.yaml +4 -1
  2. src/backend/tasks/halueval/halueval_qa.yaml +3 -0
  3. src/backend/tasks/halueval/halueval_summarization.yaml +3 -0
src/backend/tasks/halueval/halueval_dialogue.yaml CHANGED
@@ -10,11 +10,14 @@ doc_to_text: !function utils.doc_to_text_dialogue
10
  doc_to_target: !function utils.doc_to_target
11
  process_results: !function utils.process_results
12
  metric_list:
13
- - metric: em
14
  aggregation: mean
15
  higher_is_better: true
16
  - metric: correctness
17
  aggregation: mean
18
  higher_is_better: true
 
 
 
19
  metadata:
20
  - version: 0.0
 
10
  doc_to_target: !function utils.doc_to_target
11
  process_results: !function utils.process_results
12
  metric_list:
13
+ - metric: acc
14
  aggregation: mean
15
  higher_is_better: true
16
  - metric: correctness
17
  aggregation: mean
18
  higher_is_better: true
19
+ - metric: acc
20
+ aggregation: mean
21
+ higher_is_better: true
22
  metadata:
23
  - version: 0.0
src/backend/tasks/halueval/halueval_qa.yaml CHANGED
@@ -16,5 +16,8 @@ metric_list:
16
  - metric: correctness
17
  aggregation: mean
18
  higher_is_better: true
 
 
 
19
  metadata:
20
  - version: 0.0
 
16
  - metric: correctness
17
  aggregation: mean
18
  higher_is_better: true
19
+ - metric: acc
20
+ aggregation: mean
21
+ higher_is_better: true
22
  metadata:
23
  - version: 0.0
src/backend/tasks/halueval/halueval_summarization.yaml CHANGED
@@ -16,5 +16,8 @@ metric_list:
16
  - metric: correctness
17
  aggregation: mean
18
  higher_is_better: true
 
 
 
19
  metadata:
20
  - version: 0.0
 
16
  - metric: correctness
17
  aggregation: mean
18
  higher_is_better: true
19
+ - metric: acc
20
+ aggregation: mean
21
+ higher_is_better: true
22
  metadata:
23
  - version: 0.0