Source code for oumi.datasets.grpo.rewards.gsm8k_reward
# Copyright 2024 Bytedance Ltd. and/or its affiliates
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Derived from https://github.com/volcengine/verl/blob/main/verl/utils/reward_score/gsm8k.py.
This file was slightly modified to add additional args to gsm8k_reward.
"""
import re
from oumi.core.registry import RegistryType, register
def _extract_solution(solution_str, method="strict"):
"""Extract the solution from the response."""
assert method in ["strict", "flexible"]
if method == "strict":
# this also tests the formatting of the model
solutions = re.findall("#### (\\-?[0-9\\.\\,]+)", solution_str)
if len(solutions) == 0:
final_answer = None
else:
# take the last solution
final_answer = solutions[-1].replace(",", "").replace("$", "")
elif method == "flexible":
answer = re.findall("(\\-?[0-9\\.\\,]+)", solution_str)
final_answer = None
if len(answer) == 0:
# no reward is there is no answer
pass
else:
invalid_str = ["", "."]
# find the last number that is not '.'
for final_answer in reversed(answer):
if final_answer not in invalid_str:
break
return final_answer
[docs]
@register("gsm8k", RegistryType.REWARD_FUNCTION)
def gsm8k_reward(
data_source,
solution_str,
ground_truth,
extra_info,
method="strict",
format_score=0.0,
score=1.0,
):
"""The scoring function for GSM8K.
Reference: Trung, Luong, et al. "Reft: Reasoning with reinforced fine-tuning." Proceedings of the 62nd Annual
Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.
Args:
data_source: the data source
solution_str: the solution text
ground_truth: the ground truth
extra_info: extra info
method: the method to extract the solution, choices are 'strict' and 'flexible'
format_score: the score for the format
score: the score for the correct answer
""" # noqa: E501
answer = _extract_solution(solution_str=solution_str, method=method)
if answer is None:
return 0
else:
if answer == ground_truth:
return score
else:
return format_score