add bfcl utility

environment.yml

@@ -69,6 +69,7 @@ dependencies:
       - pycocotools>=2.0.8
       - vllm>=0.8.5
       - latex2sympy2>=1.9.1
+      - bfcl-eval>=2025.10.13
       - parameterized>=0.9.0
       - pyseccomp>=0.1.2
 prefix: /home/sayouse/miniconda3/envs/myenv

eureka_ml_insights/data_utils/bfcl_multiturn_utils.py

@@ -0,0 +1,57 @@
+import re, ast
+from dataclasses import dataclass
+
+import pandas as pd
+from bfcl_eval.eval_checker.multi_turn_eval.multi_turn_utils import (
+    execute_multi_turn_func_call,
+)
+
+from .transform import DFTransformBase
+
+
+@dataclass
+class BFCLMultiturnExecuteCall(DFTransformBase):
+    model_output_column: str
+    model_answer_column: str
+
+    def transform(self, df: pd.DataFrame) -> pd.DataFrame:
+        df[self.model_answer_column] = df.apply(self.execuate_model_output, axis=1)
+        return df
+
+    @staticmethod
+    def execuate_model_output(df_row):
+        """
+        Execute the model output to get the function output.
+
+        Parameters:
+            df_row (a row in dataframe): Input is a dataframe row containing model_output, initial_config, and, involved_classes.
+        Returns:
+            " ".join(execution_results) (str): A string denoting the execution of the function calls.
+        """
+        test_entry = df_row
+        response_text = test_entry["model_output"]
+        try:
+            initial_config = ast.literal_eval(test_entry["initial_config"])
+        except (ValueError, SyntaxError) as e:
+            raise ValueError(f"Invalid initial_config format: {test_entry['initial_config']}") from e
+        try:
+            involved_classes: list = eval(test_entry["involved_classes"])
+        except (ValueError, SyntaxError) as e:
+            raise ValueError(f"Invalid involved_classes format: {test_entry['involved_classes']}") from e
+        test_entry_id: str = test_entry["id"]
+        test_category: str = test_entry_id.rsplit("_", 1)[0]
+
+        func_calls = re.findall(r"\w+\([^)]*\)", response_text)
+        if len(func_calls) == 0:
+            return "No call executed"
+
+        execution_results, involved_instances = execute_multi_turn_func_call(
+            func_call_list=func_calls,
+            initial_config=initial_config,
+            involved_classes=involved_classes,
+            model_name="",
+            test_entry_id=test_entry_id,
+            long_context=("long_context" in test_category or "composite" in test_category),
+            is_evaL_run=False,
+        )
+        return " ".join(execution_results)

eureka_ml_insights/metrics/bfcl_metrics.py

@@ -0,0 +1,57 @@
+from bfcl_eval.eval_checker.multi_turn_eval.multi_turn_checker import (
+    multi_turn_checker,
+)
+
+from eureka_ml_insights.metrics.metrics_base import ClassicMetric
+
+
+class BFCLMultiturnMatch(ClassicMetric):
+    """This metric class checks if the generated function calls match the ground-truth."""
+
+    def __init__(
+        self,
+        model_output_col: str = "model_output",
+        ground_truth_col: str = "ground_truth",
+        initial_config_col: str = "initial_config",
+        involved_classes_col: str = "involved_classes",
+        test_entry_id_col: str = "id",
+    ):
+        super().__init__()
+        self.model_output_col = model_output_col
+        self.ground_truth_col = ground_truth_col
+        self.initial_config_col = initial_config_col
+        self.involved_classes_col = involved_classes_col
+        self.test_entry_id_col = test_entry_id_col
+
+    def __evaluate__(self, answer_text, target_text, initial_config, involved_classes, test_entry_id):
+        test_entry = {
+            "initial_config": eval(initial_config),
+            "involved_classes": eval(involved_classes),
+            "id": test_entry_id,
+        }
+        test_category = ""  # following the bfcl_eval's original eval code.
+        model_name = ""  # following the bfcl_eval's original eval code.
+        multi_turn_ground_truth_list = eval(target_text)
+
+        accuracy_checker_result = multi_turn_checker(
+            answer_text,
+            multi_turn_ground_truth_list,
+            test_entry,
+            test_category,
+            model_name,
+        )
+        return str(accuracy_checker_result["valid"])
+
+    def evaluate(self, data):
+        self.validate_data(data)
+        data[self.__class__.__name__ + "_result"] = data.apply(
+            lambda x: self.__evaluate__(
+                x[self.model_output_col],
+                x[self.ground_truth_col],
+                x[self.initial_config_col],
+                x[self.involved_classes_col],
+                x[self.test_entry_id_col],
+            ),
+            axis=1,
+        )
+        return data

setup.py

@@ -1,5 +1,4 @@
-from setuptools import setup, find_packages
-
+from setuptools import find_packages, setup
 
 setup(
     name='eureka_ml_insights',
@@ -44,6 +43,7 @@
         'pycocotools>=2.0.8',
         'vllm>=0.8.5',
         'latex2sympy2>=1.9.1',
+        'bfcl-eval>=2025.10.13',
         'parameterized>=0.9.0',
         'pyseccomp>=0.1.2',
     ],
@@ -54,9 +54,9 @@
     },
     classifiers=[
         # Full list at https://pypi.org/classifiers/
-        'Programming Language :: Python :: 3',
-        'License :: Apache License 2.0',
-        'Operating System :: OS Independent',
+        "Programming Language :: Python :: 3",
+        "License :: Apache License 2.0",
+        "Operating System :: OS Independent",
     ],
-    python_requires='>=3.8',
+    python_requires=">=3.8",
 )