Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
show perplexity for 5-shot too
Browse files
src/leaderboard/read_evals.py
CHANGED
@@ -95,9 +95,13 @@ class EvalResult:
|
|
95 |
for task in Tasks:
|
96 |
task = task.value
|
97 |
|
|
|
|
|
|
|
|
|
98 |
# We average all scores of a given metric (not all metrics are present in all files)
|
99 |
accs = np.array([v.get(task.metric, None) for k, v in data["results"].items() if
|
100 |
-
task.benchmark == k and n_shot.get(k, -1) ==
|
101 |
if accs.size == 0 or any([acc is None for acc in accs]):
|
102 |
continue
|
103 |
|
|
|
95 |
for task in Tasks:
|
96 |
task = task.value
|
97 |
|
98 |
+
task_n_shot_num = n_shot_num
|
99 |
+
if 'perplexity' in task.metric: # perplexity is the same for 0-shot and 5-shot and is calculated only with 0-shot
|
100 |
+
task_n_shot_num = 0
|
101 |
+
|
102 |
# We average all scores of a given metric (not all metrics are present in all files)
|
103 |
accs = np.array([v.get(task.metric, None) for k, v in data["results"].items() if
|
104 |
+
task.benchmark == k and n_shot.get(k, -1) == task_n_shot_num])
|
105 |
if accs.size == 0 or any([acc is None for acc in accs]):
|
106 |
continue
|
107 |
|