File size: 4,703 Bytes
d65b589 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 | # 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.
import re
from typing import Any
# Constants for normalization
SUBSTITUTIONS = [
("an ", ""),
("a ", ""),
(".$", "$"),
("\\$", ""),
(r"\ ", ""),
(" ", ""),
("mbox", "text"),
(",\\text{and}", ","),
("\\text{and}", ","),
("\\text{m}", "\\text{}"),
]
REMOVED_EXPRESSIONS = [
"square",
"ways",
"integers",
"dollars",
"mph",
"inches",
"hours",
"km",
"units",
"\\ldots",
"sue",
"points",
"feet",
"minutes",
"digits",
"cents",
"degrees",
"cm",
"gm",
"pounds",
"meters",
"meals",
"edges",
"students",
"childrentickets",
"multiples",
"\\text{s}",
"\\text{.}",
"\\text{\ns}",
"\\text{}^2",
"\\text{}^3",
"\\text{\n}",
"\\text{}",
r"\mathrm{th}",
r"^\circ",
r"^{\circ}",
r"\;",
r",\!",
"{,}",
'"',
"\\dots",
]
def normalize_final_answer(final_answer: str) -> str:
"""Normalize a final answer to a quantitative reasoning question.
Args:
final_answer: The answer string to normalize
Returns:
Normalized answer string
"""
final_answer = final_answer.split("=")[-1]
# Apply substitutions and removals
for before, after in SUBSTITUTIONS:
final_answer = final_answer.replace(before, after)
for expr in REMOVED_EXPRESSIONS:
final_answer = final_answer.replace(expr, "")
# Extract and normalize LaTeX math
final_answer = re.sub(r"(.*?)(\$)(.*?)(\$)(.*)", "$\\3$", final_answer)
final_answer = re.sub(r"(\\text\{)(.*?)(\})", "\\2", final_answer)
final_answer = re.sub(r"(\\textbf\{)(.*?)(\})", "\\2", final_answer)
final_answer = re.sub(r"(\\overline\{)(.*?)(\})", "\\2", final_answer)
final_answer = re.sub(r"(\\boxed\{)(.*)(\})", "\\2", final_answer)
# Normalize shorthand TeX:
# \fracab -> \frac{a}{b}
# \frac{abc}{bef} -> \frac{abc}{bef}
# \fracabc -> \frac{a}{b}c
# \sqrta -> \sqrt{a}
# \sqrtab -> sqrt{a}b
final_answer = re.sub(r"(frac)([^{])(.)", "frac{\\2}{\\3}", final_answer)
final_answer = re.sub(r"(sqrt)([^{])", "sqrt{\\2}", final_answer)
final_answer = final_answer.replace("$", "")
# Normalize numbers
if final_answer.replace(",", "").isdigit():
final_answer = final_answer.replace(",", "")
return final_answer.strip()
def accuracy_reward(response: str, ground_truth: str) -> float:
match = re.findall(r"(?i)Answer\s*:\s*([^\n]+)", response)
answer = match[-1] if match else "[INVALID]"
if normalize_final_answer(answer) == normalize_final_answer(ground_truth):
return 1.0
else:
return -1.0
def soft_overlong_punishment(response_length: int, max_response_length: int, overlong_buffer_length: int):
expected_len = max_response_length - overlong_buffer_length
if response_length <= expected_len:
return 0.0
elif response_length <= max_response_length:
return (expected_len - response_length) / overlong_buffer_length
else:
return -1.0
def compute_score(
reward_inputs: list[dict[str, Any]],
max_response_length: int,
overlong_buffer_length: int,
overlong_penalty_factor: float,
) -> list[dict[str, float]]:
if not isinstance(reward_inputs, list):
raise ValueError("Please use `reward_type=batch` for dapo reward function.")
scores = []
for reward_input in reward_inputs:
response = reward_input["response"][-300:] # The longest answer in MATH-500 has 159 characters
accuracy_score = accuracy_reward(response, reward_input["ground_truth"])
overlong_score = soft_overlong_punishment(
reward_input["response_length"], max_response_length, overlong_buffer_length
)
scores.append(
{
"overall": accuracy_score + overlong_score * overlong_penalty_factor,
"accuracy": accuracy_score,
"overlong": overlong_score,
"accuracy_normalized": 0.5 * (accuracy_score + 1.0),
}
)
return scores
|