Standardize metrics

lm_eval/api/registry.py

@@ -1,19 +1,10 @@
-<<<<<<< HEAD
 import os
 import logging
 import evaluate
 import collections
 from functools import partial
 
 from lm_eval.api.model import LM
-=======
-import logging
-
-import evaluate
-
-from lm_eval.api.model import LM
-
->>>>>>> 4d10ad56b1ffe569467eee2297e2317c99313118
 
 eval_logger = logging.getLogger("lm-eval")
 
@@ -129,7 +120,6 @@ def decorate(fn):
     return decorate
 
 
-<<<<<<< HEAD
 def get_metric(name):
 
     if name in METRIC_REGISTRY:
@@ -139,17 +129,6 @@ def get_metric(name):
 
 
 def get_evaluate(name, **kwargs):
-=======
-def get_metric(name, hf_evaluate_metric=False):
-    if not hf_evaluate_metric:
-        if name in METRIC_REGISTRY:
-            return METRIC_REGISTRY[name]
-        else:
-            eval_logger.warning(
-                f"Could not find registered metric '{name}' in lm-eval, searching in HF Evaluate library..."
-            )
->>>>>>> 4d10ad56b1ffe569467eee2297e2317c99313118
-
     try:
 
         class HFEvaluateAdaptor:

lm_eval/api/task.py

@@ -19,6 +19,9 @@
     mean,
     weighted_perplexity,
 <<<<<<< HEAD
+<<<<<<< HEAD
+=======
+>>>>>>> cda25fef4e1df2f4bc2dab3ec6668ae9f5bf7296
     bits_per_byte,
 )
 from lm_eval.api.registry import (
@@ -27,6 +30,7 @@
     get_aggregation,
     METRIC_REGISTRY,
     DEFAULT_METRIC_REGISTRY,
+<<<<<<< HEAD
 =======
 )
 from lm_eval.api.registry import (
@@ -37,6 +41,8 @@
     get_metric_aggregation,
     is_higher_better,
 >>>>>>> 4d10ad56b1ffe569467eee2297e2317c99313118
+=======
+>>>>>>> cda25fef4e1df2f4bc2dab3ec6668ae9f5bf7296
 )
 from lm_eval.filters import build_filter_ensemble
 from lm_eval.prompts import get_prompt