modified import origin

lm_eval/evaluator.py

@@ -9,7 +9,7 @@ import numpy as np
 import lm_eval.api
 import lm_eval.tasks
 import lm_eval.models
-import lm_eval.metrics
+import lm_eval.api.metrics
 
 from lm_eval.utils import (
     positional_deprecated,
@@ -280,7 +280,7 @@ def evaluate(
             # hotfix: bleu, chrf, ter seem to be really expensive to bootstrap
             # so we run them less iterations. still looking for a cleaner way to do this
 
-            stderr = lm_eval.metrics.stderr_for_metric(
+            stderr = lm_eval.api.metrics.stderr_for_metric(
                 metric=task.aggregation()[metric],
                 bootstrap_iters=min(bootstrap_iters, 1000)
                 if metric in ["bleu", "chrf", "ter"]