feat: add gsm8k parser
Browse files- llmdataparser/gsm8k_parser.py +93 -0
- tests/test_gsm8k_parser.py +183 -0
llmdataparser/gsm8k_parser.py
ADDED
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from dataclasses import dataclass
|
2 |
+
from typing import Any, ClassVar
|
3 |
+
|
4 |
+
from llmdataparser.base_parser import HuggingFaceDatasetParser, HuggingFaceParseEntry
|
5 |
+
from llmdataparser.prompts import GSM8K_SYSTEM_PROMPT
|
6 |
+
|
7 |
+
|
8 |
+
@dataclass(frozen=True, kw_only=True, slots=True)
|
9 |
+
class GSM8KParseEntry(HuggingFaceParseEntry):
|
10 |
+
"""Custom entry class for GSM8K, with fields specific to this dataset parser."""
|
11 |
+
|
12 |
+
solution: str
|
13 |
+
numerical_answer: int | float
|
14 |
+
task_name: str
|
15 |
+
|
16 |
+
@classmethod
|
17 |
+
def create(
|
18 |
+
cls,
|
19 |
+
prompt: str,
|
20 |
+
answer: str,
|
21 |
+
raw_question: str,
|
22 |
+
raw_answer: str,
|
23 |
+
solution: str,
|
24 |
+
numerical_answer: int | float,
|
25 |
+
task_name: str,
|
26 |
+
) -> "GSM8KParseEntry":
|
27 |
+
return cls(
|
28 |
+
prompt=prompt,
|
29 |
+
answer=answer,
|
30 |
+
raw_question=raw_question,
|
31 |
+
raw_answer=raw_answer,
|
32 |
+
solution=solution,
|
33 |
+
numerical_answer=numerical_answer,
|
34 |
+
task_name=task_name,
|
35 |
+
)
|
36 |
+
|
37 |
+
|
38 |
+
class GSM8KDatasetParser(HuggingFaceDatasetParser[GSM8KParseEntry]):
|
39 |
+
"""Parser for the GSM8K dataset."""
|
40 |
+
|
41 |
+
_data_source: ClassVar[str] = "openai/gsm8k"
|
42 |
+
_task_names: ClassVar[list[str]] = ["main", "socratic"]
|
43 |
+
_default_task: ClassVar[str] = "main"
|
44 |
+
_default_system_prompt: ClassVar[str] = GSM8K_SYSTEM_PROMPT
|
45 |
+
|
46 |
+
def process_entry(
|
47 |
+
self, row: dict[str, Any], task_name: str | None = None, **kwargs: Any
|
48 |
+
) -> GSM8KParseEntry:
|
49 |
+
"""Process a single GSM8K entry."""
|
50 |
+
task = task_name or self._get_current_task(row)
|
51 |
+
raw_question = row["question"]
|
52 |
+
raw_answer = row["answer"]
|
53 |
+
|
54 |
+
# Extract numerical answer (always after '####' in GSM8K)
|
55 |
+
numerical_str = raw_answer.split("####")[-1].strip().replace(",", "")
|
56 |
+
# Convert string to number
|
57 |
+
try:
|
58 |
+
numerical_answer = float(numerical_str)
|
59 |
+
if numerical_answer.is_integer():
|
60 |
+
numerical_answer = int(numerical_answer)
|
61 |
+
except ValueError:
|
62 |
+
raise ValueError(f"Could not convert '{numerical_str}' to number")
|
63 |
+
|
64 |
+
# Extract solution (everything before '####')
|
65 |
+
solution = raw_answer.split("####")[0].strip()
|
66 |
+
|
67 |
+
prompt = f"{self._system_prompt}\n{raw_question}"
|
68 |
+
|
69 |
+
return GSM8KParseEntry.create(
|
70 |
+
prompt=prompt,
|
71 |
+
answer=str(numerical_answer),
|
72 |
+
raw_question=raw_question,
|
73 |
+
raw_answer=raw_answer,
|
74 |
+
solution=solution,
|
75 |
+
numerical_answer=numerical_answer, # Now guaranteed to be int or float
|
76 |
+
task_name=task, # Guarantee non-None
|
77 |
+
)
|
78 |
+
|
79 |
+
|
80 |
+
if __name__ == "__main__":
|
81 |
+
from pprint import pprint
|
82 |
+
|
83 |
+
parser = GSM8KDatasetParser()
|
84 |
+
parser.load()
|
85 |
+
parser.parse()
|
86 |
+
|
87 |
+
parsed_data = parser.get_parsed_data
|
88 |
+
pprint(parsed_data[0].prompt)
|
89 |
+
pprint(parsed_data[0].answer)
|
90 |
+
pprint(parsed_data[0].raw_question)
|
91 |
+
pprint(parsed_data[0].raw_answer)
|
92 |
+
pprint(parsed_data[0].solution)
|
93 |
+
pprint(parsed_data[0].numerical_answer)
|
tests/test_gsm8k_parser.py
ADDED
@@ -0,0 +1,183 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pytest
|
2 |
+
|
3 |
+
from llmdataparser.gsm8k_parser import GSM8KDatasetParser, GSM8KParseEntry
|
4 |
+
|
5 |
+
|
6 |
+
@pytest.fixture
|
7 |
+
def gsm8k_parser():
|
8 |
+
"""Create a GSM8K parser instance for testing."""
|
9 |
+
return GSM8KDatasetParser()
|
10 |
+
|
11 |
+
|
12 |
+
@pytest.fixture
|
13 |
+
def loaded_gsm8k_parser(gsm8k_parser):
|
14 |
+
"""Create and load a GSM8K parser instance for testing."""
|
15 |
+
gsm8k_parser.load(
|
16 |
+
task_name="main", split="test"
|
17 |
+
) # Using test split as it's smaller
|
18 |
+
return gsm8k_parser
|
19 |
+
|
20 |
+
|
21 |
+
@pytest.fixture
|
22 |
+
def sample_row():
|
23 |
+
"""Create a sample GSM8K data row for testing."""
|
24 |
+
return {
|
25 |
+
"question": "Janet has 3 apples. She buys 2 more. How many apples does she have now?",
|
26 |
+
"answer": "Let's solve this step by step:\n1) Initially, Janet has 3 apples\n2) She buys 2 more apples\n3) Total apples = 3 + 2\n#### 5",
|
27 |
+
}
|
28 |
+
|
29 |
+
|
30 |
+
def test_gsm8k_parse_entry_creation_valid():
|
31 |
+
"""Test valid creation of GSM8KParseEntry."""
|
32 |
+
entry = GSM8KParseEntry.create(
|
33 |
+
prompt="Test prompt",
|
34 |
+
answer="5",
|
35 |
+
raw_question="Test question",
|
36 |
+
raw_answer="Solution steps #### 5",
|
37 |
+
solution="Solution steps",
|
38 |
+
task_name="main",
|
39 |
+
numerical_answer=5,
|
40 |
+
)
|
41 |
+
assert isinstance(entry, GSM8KParseEntry)
|
42 |
+
assert entry.prompt == "Test prompt"
|
43 |
+
assert entry.answer == "5"
|
44 |
+
assert entry.solution == "Solution steps"
|
45 |
+
assert entry.numerical_answer == 5
|
46 |
+
assert entry.task_name == "main"
|
47 |
+
|
48 |
+
|
49 |
+
def test_gsm8k_parser_initialization(gsm8k_parser):
|
50 |
+
"""Test GSM8K parser initialization."""
|
51 |
+
assert gsm8k_parser._data_source == "openai/gsm8k"
|
52 |
+
assert gsm8k_parser._default_task == "main"
|
53 |
+
assert gsm8k_parser._task_names == ["main", "socratic"]
|
54 |
+
assert (
|
55 |
+
gsm8k_parser.get_huggingface_link
|
56 |
+
== "https://huggingface.co/datasets/openai/gsm8k"
|
57 |
+
)
|
58 |
+
|
59 |
+
|
60 |
+
def test_load_dataset(loaded_gsm8k_parser):
|
61 |
+
"""Test loading the dataset."""
|
62 |
+
assert loaded_gsm8k_parser.raw_data is not None
|
63 |
+
assert loaded_gsm8k_parser.split_names == [
|
64 |
+
"test"
|
65 |
+
] # Since we specifically loaded the test split
|
66 |
+
assert loaded_gsm8k_parser._current_task == "main"
|
67 |
+
|
68 |
+
|
69 |
+
@pytest.mark.integration
|
70 |
+
def test_full_parse_workflow(loaded_gsm8k_parser):
|
71 |
+
"""Test the complete workflow of loading and parsing data."""
|
72 |
+
# Parse the test split
|
73 |
+
loaded_gsm8k_parser.parse(split_names="test", force=True)
|
74 |
+
parsed_data = loaded_gsm8k_parser.get_parsed_data
|
75 |
+
|
76 |
+
# Basic checks
|
77 |
+
assert len(parsed_data) > 0
|
78 |
+
|
79 |
+
# Check first entry structure
|
80 |
+
first_entry = parsed_data[0]
|
81 |
+
assert isinstance(first_entry, GSM8KParseEntry)
|
82 |
+
assert first_entry.task_name == "main"
|
83 |
+
assert isinstance(first_entry.numerical_answer, (str, int, float))
|
84 |
+
assert "####" in first_entry.raw_answer
|
85 |
+
assert first_entry.solution
|
86 |
+
assert first_entry.prompt.startswith(loaded_gsm8k_parser._system_prompt)
|
87 |
+
|
88 |
+
|
89 |
+
def test_process_entry(gsm8k_parser, sample_row):
|
90 |
+
"""Test processing of a single GSM8K entry."""
|
91 |
+
entry = gsm8k_parser.process_entry(sample_row, task_name="main")
|
92 |
+
|
93 |
+
assert isinstance(entry, GSM8KParseEntry)
|
94 |
+
assert entry.numerical_answer == 5
|
95 |
+
assert "Janet has 3 apples" in entry.raw_question
|
96 |
+
assert "#### 5" in entry.raw_answer
|
97 |
+
assert "Let's solve this step by step:" in entry.solution
|
98 |
+
assert gsm8k_parser._system_prompt in entry.prompt
|
99 |
+
assert entry.task_name == "main"
|
100 |
+
|
101 |
+
|
102 |
+
@pytest.mark.parametrize("split_name", ["invalid_split", "wrong_split"])
|
103 |
+
def test_parse_with_invalid_split(gsm8k_parser, split_name):
|
104 |
+
"""Test parsing with invalid split names."""
|
105 |
+
gsm8k_parser.raw_data = {"train": [], "test": []} # Mock data
|
106 |
+
|
107 |
+
with pytest.raises(
|
108 |
+
ValueError, match=f"Split '{split_name}' not found in the dataset"
|
109 |
+
):
|
110 |
+
gsm8k_parser.parse(split_name)
|
111 |
+
|
112 |
+
|
113 |
+
def test_parse_without_loaded_data(gsm8k_parser):
|
114 |
+
"""Test parsing without loading data first."""
|
115 |
+
with pytest.raises(
|
116 |
+
ValueError, match="No data loaded. Please load the dataset first"
|
117 |
+
):
|
118 |
+
gsm8k_parser.parse()
|
119 |
+
|
120 |
+
|
121 |
+
@pytest.mark.parametrize(
|
122 |
+
"test_case",
|
123 |
+
[
|
124 |
+
{"question": "Test question", "answer": "Some solution steps #### 42"},
|
125 |
+
{
|
126 |
+
"question": "Test question",
|
127 |
+
"answer": "Complex solution\nWith multiple lines\n#### 123.45",
|
128 |
+
},
|
129 |
+
{"question": "Test question", "answer": "No steps #### 0"},
|
130 |
+
],
|
131 |
+
)
|
132 |
+
def test_numerical_answer_extraction(gsm8k_parser, test_case):
|
133 |
+
"""Test extraction of numerical answers from different formats."""
|
134 |
+
entry = gsm8k_parser.process_entry(test_case, task_name="main")
|
135 |
+
assert str(entry.numerical_answer) == test_case["answer"].split("####")[
|
136 |
+
-1
|
137 |
+
].strip().replace(",", "")
|
138 |
+
|
139 |
+
|
140 |
+
def test_solution_extraction(gsm8k_parser):
|
141 |
+
"""Test extraction of solution steps."""
|
142 |
+
row = {
|
143 |
+
"question": "Test question",
|
144 |
+
"answer": "Step 1: Do this\nStep 2: Do that\n#### 42",
|
145 |
+
}
|
146 |
+
|
147 |
+
entry = gsm8k_parser.process_entry(row, task_name="main")
|
148 |
+
assert entry.solution == "Step 1: Do this\nStep 2: Do that"
|
149 |
+
assert entry.task_name == "main"
|
150 |
+
assert "####" not in entry.solution
|
151 |
+
|
152 |
+
|
153 |
+
def test_parser_properties(gsm8k_parser):
|
154 |
+
"""Test parser property getters."""
|
155 |
+
assert gsm8k_parser.task_names == ["main", "socratic"]
|
156 |
+
assert gsm8k_parser.total_tasks == 2
|
157 |
+
|
158 |
+
|
159 |
+
def test_parser_string_representation(loaded_gsm8k_parser):
|
160 |
+
"""Test string representation of parser."""
|
161 |
+
repr_str = str(loaded_gsm8k_parser)
|
162 |
+
assert "GSM8KDatasetParser" in repr_str
|
163 |
+
assert "openai/gsm8k" in repr_str
|
164 |
+
assert "main" in repr_str
|
165 |
+
assert "loaded" in repr_str
|
166 |
+
|
167 |
+
|
168 |
+
@pytest.mark.integration
|
169 |
+
def test_different_splits_parsing(gsm8k_parser):
|
170 |
+
"""Test parsing different splits of the dataset."""
|
171 |
+
# Load and parse test split
|
172 |
+
gsm8k_parser.load(task_name="main", split="test")
|
173 |
+
gsm8k_parser.parse(split_names="test", force=True)
|
174 |
+
test_count = len(gsm8k_parser.get_parsed_data)
|
175 |
+
|
176 |
+
# Load and parse train split
|
177 |
+
gsm8k_parser.load(task_name="main", split="train")
|
178 |
+
gsm8k_parser.parse(split_names="train", force=True)
|
179 |
+
train_count = len(gsm8k_parser.get_parsed_data)
|
180 |
+
|
181 |
+
assert test_count > 0
|
182 |
+
assert train_count > 0
|
183 |
+
assert train_count != test_count
|