Filters bugfix; add `metrics` and `filter` to logged sample (#2517)

* allow !function filters

* bugfix

* nit

* add `filter` to logged samples

* add `filter` and `metric` to logged samples to identification

* convert `metric` to `metrics`: list

lm_eval/api/registry.py

 import logging
-from typing import Callable, Dict
+from typing import Callable, Dict, Union
 
 import evaluate as hf_evaluate
 
@@ -185,8 +185,12 @@ def register_filter(name):
     return decorate
 
 
-def get_filter(filter_name: str) -> type:
+def get_filter(filter_name: Union[str, Callable]) -> Callable:
     try:
         return FILTER_REGISTRY[filter_name]
-    except KeyError:
-        eval_logger.warning(f"filter `{filter_name}` is not registered!")
+    except KeyError as e:
+        if callable(filter_name):
+            return filter_name
+        else:
+            eval_logger.warning(f"filter `{filter_name}` is not registered!")
+            raise e

lm_eval/evaluator.py

@@ -552,6 +552,8 @@ def evaluate(
                         "filtered_resps": [
                             req.filtered_resps[filter_key] for req in requests
                         ],
+                        "filter": filter_key,
+                        "metrics": list(metrics.keys()),
                         "doc_hash": hash_string(
                             json.dumps(
                                 requests[0].doc,

lm_eval/filters/extraction.py

@@ -37,16 +37,18 @@ class RegexFilter(Filter):
                 if match:
                     match = match[self.group_select]
                     if isinstance(match, tuple):
-                        match = [m for m in match if m][0]
+                        match = [m for m in match if m]
+                        if match:
+                            match = match[0]
+                        else:
+                            match = self.fallback
                     match = match.strip()
                 else:
                     match = self.fallback
                 filtered.append(match)
             return filtered
 
-        # print(resps)
         filtered_resps = list(map(lambda x: filter_set(x), resps))
-        # print(filtered_resps)
 
         return filtered_resps