forked from eval-protocol/python-sdk
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtest_math.py
More file actions
540 lines (463 loc) · 26.7 KB
/
Copy pathtest_math.py
File metadata and controls
540 lines (463 loc) · 26.7 KB
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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
"""
Tests for math reward functions.
"""
import re
from typing import List, Optional, Tuple, Union # Added Tuple
import pytest
# Removed: from eval_protocol.rewards.advanced_math import advanced_math_reward
from eval_protocol.models import (
EvaluateResult,
Message, # Added Message import
MetricResult,
)
from eval_protocol.rewards.math import compare_numbers, extract_numbers, math_reward
class TestExtractNumbers:
# --- PRIORITY 1: Boxed LaTeX ---
def test_P1_boxed_simple_number(self):
text = "The answer is \\boxed{42}."
expected = [("\\boxed{42}", 42.0)]
assert extract_numbers(text) == expected
def test_P1_boxed_simple_fraction_string(self):
text = "The answer is \\boxed{1/2}."
expected = [("\\boxed{1/2}", 0.5)]
assert extract_numbers(text) == expected
def test_P1_boxed_latex_fraction(self):
text = "The answer is \\boxed{\\frac{3}{4}}."
expected = [("\\boxed{\\frac{3}{4}}", 0.75)]
assert extract_numbers(text) == expected
def test_P1_boxed_mcq_letter(self):
text = "The answer is \\boxed{A}."
expected = [("\\boxed{A}", "A")]
assert extract_numbers(text) == expected
def test_P1_boxed_or_expression(self):
text = "The roots are \\boxed{x=1 \\text{ or } x=2}."
expected = [("\\boxed{x=1 \\text{ or } x=2}", "x=1 \\text{ or } x=2")]
assert extract_numbers(text) == expected
def test_P1_boxed_complex_content_as_string_or_ignored(self):
text_no_or = "The answer is \\boxed{x^2+1}."
# P1: found_any_boxed_expr = True. boxed_answers is empty. Returns []. Correct.
assert extract_numbers(text_no_or) == []
text_with_or = "The answer is \\boxed{x^2+1 \\text{ or } y=0}."
expected_with_or = [("\\boxed{x^2+1 \\text{ or } y=0}", "x^2+1 \\text{ or } y=0")]
assert extract_numbers(text_with_or) == expected_with_or
def test_P1_multiple_boxed_expressions(self):
text = "Answers: \\boxed{1}, then \\boxed{B}, and finally \\boxed{1/2 or 3/2}"
# Order of extraction depends on finditer, so check presence and content
results = extract_numbers(text)
assert len(results) == 3
assert ("\\boxed{1}", 1.0) in results
assert ("\\boxed{B}", "B") in results
assert ("\\boxed{1/2 or 3/2}", "1/2 or 3/2") in results
# --- PRIORITY 2: GSM8K-style (#### ...) ---
def test_P2_gsm8k_simple_number(self):
text = "The final answer is #### 123." # Input with period
expected = [("#### 123", 123.0)] # Regex for num content won't include trailing period in m.group(0)
assert extract_numbers(text) == expected
def test_P2_gsm8k_with_commas_and_decimal(self):
text = "The final answer is #### 1,234.56." # Input with period
expected = [("#### 1,234.56", 1234.56)] # Expect m.group(0) to be "#### 1,234.56"
assert extract_numbers(text) == expected
def test_P2_gsm8k_negative_number(self):
text = "The final answer is #### -7." # Input with period
expected = [("#### -7", -7.0)]
assert extract_numbers(text) == expected
def test_P2_gsm8k_takes_priority_over_mcq_and_general(self):
text = "Answer is (A). The number is 10. Final: #### 100." # Input with period
expected = [("#### 100", 100.0)]
assert extract_numbers(text) == expected
def test_P2_multiple_gsm8k_markers(self):
text = "First #### 1. Second #### 2." # Input with periods
results = extract_numbers(text)
assert len(results) == 2
assert ("#### 1", 1.0) in results
assert ("#### 2", 2.0) in results
# --- PRIORITY 3: Multiple Choice Question (MCQ) ---
# MCQ extraction is now handled by multiple_choice_math_reward.py
# These tests verify that extract_numbers from math.py NO LONGER extracts MCQs.
def test_P3_mcq_various_formats(self):
assert extract_numbers("(A)") == []
assert extract_numbers("B.") == []
assert extract_numbers("[C]") == []
assert extract_numbers("{D}") == []
assert extract_numbers(" E ") == []
def test_P3_mcq_multiple_different_mcqs(self):
text = "Options: (A), B., [C]"
results = extract_numbers(text)
assert results == []
def test_P3_mcq_takes_priority_over_general_numbers(self):
# Since MCQ is no longer extracted by math.extract_numbers, general numbers should be found.
text = "The number is 10, but the choice is (A)."
expected = [("10", 10.0)]
assert extract_numbers(text) == expected
# --- PRIORITY 4: General Fallback Extraction ---
def test_P4_general_integers(self):
text = "The answer is 42. Another value is -17."
results = extract_numbers(text)
assert ("42", 42.0) in results
assert ("-17", -17.0) in results
assert len(results) == 2
def test_P4_general_decimals(self):
text = "The value of pi is 3.14159."
expected = [("3.14159", 3.14159)]
assert extract_numbers(text) == expected
def test_P4_general_scientific_notation(self):
text = "Avogadro's number is approximately 6.022e23."
expected = [("6.022e23", 6.022e23)]
assert extract_numbers(text) == expected
def test_P4_general_fractions_plain(self):
text = "One half is 1/2 and three quarters is 3/4."
results = extract_numbers(text)
# m.group(0) for "1/2 and" with the current frac_pattern should be "1/2"
assert ("1/2", 0.5) in results
assert ("3/4", 0.75) in results
assert len(results) == 2
def test_P4_general_latex_fraction_outside_box(self):
text = "Answer: $\\frac{1}{2}$"
expected = [("\\frac{1}{2}", 0.5)]
assert extract_numbers(text) == expected
def test_P4_general_latex_sci_notation_outside_box(self):
text = "Value: $3 \\times 10^{8}$"
expected = [("3 \\times 10^{8}", 3e8)]
assert extract_numbers(text) == expected
def test_P4_general_numbers_with_units_no_algebraic_vars(self):
text = "The distance is 42 km and the weight is 3.5 kg."
results = extract_numbers(text)
# Order might vary, check presence
assert ("42 km", 42.0) in results
assert ("3.5 kg", 3.5) in results
assert len(results) == 2
def test_P4_general_avoids_coefficients_plain_text(self):
assert extract_numbers("4y") == []
assert extract_numbers("4 y") == []
results = extract_numbers("answer = 4x + 2")
assert len(results) == 1
assert results[0] == ("2", 2.0) # Only "2" should be extracted
def test_P4_general_avoids_coefficients_latex(self):
assert extract_numbers("$4y$") == []
assert extract_numbers("$4 y$") == []
assert extract_numbers("$x=4a+2b$") == []
results_eq = extract_numbers("$x=4$") # "4" is extracted from "$x=4$"
assert len(results_eq) == 1
assert results_eq[0] == ("4", 4.0)
def test_P4_general_multiple_formats_fallback(self):
text = "Values: 42, then 3.14, also $1/4$, and $10 \\text{ m}$, finally $5.5 \\times 10^{6} \\text{ Hz}$"
results = extract_numbers(text)
extracted_values = {r[1] for r in results}
expected_values = {42.0, 3.14, 0.25, 10.0, 5.5e6}
assert extracted_values == expected_values
# Check original texts for some to ensure they are reasonable
# This can be tricky due to overlapping regexes and filtering in fallback.
# For example, "$10 \\text{ m}$" might extract "10" with original text "10".
# For "$1/4$", original text should be "1/4".
# For "$5.5 \\times 10^{6} \\text{ Hz}$", original text "5.5 \\times 10^{6}".
assert any(r[0] == "1/4" and r[1] == 0.25 for r in results)
assert any(r[0] == "5.5 \\times 10^{6}" and r[1] == 5.5e6 for r in results)
# --- Overall Priority Tests ---
def test_priority_boxed_over_all_else(self):
text = "The answer is (A), also 10, and #### 20, but finally \\boxed{30}."
expected = [("\\boxed{30}", 30.0)]
assert extract_numbers(text) == expected
def test_priority_gsm8k_over_mcq_and_general(self):
text = "The answer is (A), also 10, and finally #### 20." # Period after 20
expected = [("#### 20", 20.0)]
assert extract_numbers(text) == expected
def test_priority_mcq_over_general(self):
# MCQ logic removed, so general number '10' should be extracted.
text = "The answer is 10, but choose (A)."
expected = [("10", 10.0)]
assert extract_numbers(text) == expected
def test_no_answer_found(self):
text = "This is just some text without any clear answer."
assert extract_numbers(text) == []
def test_empty_string(self):
assert extract_numbers("") == []
def test_extract_numbers_issue_3_scenario_recheck(self):
text_coeff = "This is $4 y$."
assert extract_numbers(text_coeff) == []
text_valid_num_in_latex = "This is $v_R = 4 \\mathrm{km/h}$."
results_valid = extract_numbers(text_valid_num_in_latex)
assert len(results_valid) == 1
assert results_valid[0] == ("4", 4.0)
text_combined = "Solution with $4 y$ and also $v_R = 4 \\mathrm{km/h}$."
results_combined = extract_numbers(text_combined)
assert len(results_combined) == 1
assert results_combined[0] == ("4", 4.0)
# --- PRIORITY 0: HTML/XML tag–wrapped numeric answers ---
def test_P0_html_tag_simple_number(self):
text = "Some text <answer>77</answer> more text."
expected = [("<answer>77</answer>", 77.0)]
assert extract_numbers(text) == expected
def test_P0_html_tag_with_whitespace(self):
text = "<answer> 3.14 </answer>"
expected = [("<answer> 3.14 </answer>", 3.14)]
assert extract_numbers(text) == expected
def test_P0_html_tag_string_content(self):
text = "<answer>Not a number</answer>"
expected: List[Tuple[str, Union[str, float]]] = []
assert extract_numbers(text) == expected
def test_P0_html_tag_multiple_answer_tags(self):
text = "<answer>1</answer> and <answer>2.5</answer>"
results = extract_numbers(text)
assert ("<answer>1</answer>", 1.0) in results
assert ("<answer>2.5</answer>", 2.5) in results
assert len(results) == 2
def test_P0_html_tag_priority_over_general(self):
text = "The answer is <answer>42</answer> but also 99."
expected = [("<answer>42</answer>", 42.0)]
assert extract_numbers(text) == expected
class TestCompareNumbers:
def test_exact_match(self):
is_match, similarity = compare_numbers(42.0, 42.0)
assert is_match is True
assert similarity == 1.0
def test_close_match(self):
is_match, similarity = compare_numbers(3.14159, 3.14, relative_tolerance=0.01)
assert is_match is True
assert similarity == 1.0
def test_not_close_match(self):
is_match, similarity = compare_numbers(10.0, 11.0, relative_tolerance=0.01)
assert is_match is False
assert similarity < 1.0
def test_very_different(self):
is_match, similarity = compare_numbers(100.0, 200.0, relative_tolerance=0.01)
assert is_match is False
assert similarity == 0.0
def test_zero_expected(self):
is_match, similarity = compare_numbers(0.0, 0.00001, absolute_tolerance=0.0001)
assert is_match is True
assert similarity == 1.0
is_match, similarity = compare_numbers(0.0, 0.0001, absolute_tolerance=0.0001)
assert is_match is True
assert similarity == 1.0
is_match, similarity = compare_numbers(0.0, 0.001, absolute_tolerance=0.0001)
assert is_match is False
assert similarity < 1.0
class TestMathReward:
# Helper to create a common prompt
def _get_prompt(
self,
) -> Optional[List[Message]]: # Ensure return type matches Optional if used as such
return [Message(role="user", content="Solve the math problem.")]
def test_basic_match_boxed(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="It is \\boxed{4}.")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Answer is \\boxed{4}."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 1.0
# assert result.score == 1.0 # Corrected access
def test_basic_match_gsm8k(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="The result is #### 4")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Final answer: #### 4"
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 1.0
# assert result.score == 1.0 # Corrected access
def test_basic_match_mcq(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="My answer is (A).")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Choice (A)."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
# assert result.score == 0.0 # Duplicate line removed
assert (
result.reason is not None
and "Could not extract answers from original message (ground truth)" in result.reason
)
def test_type_mismatch_mcq_vs_number(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="It is 4.")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "The number is 4."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 1.0
# assert result.score == 1.0 # Corrected access
def test_close_match_tolerance(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="Pi is \\boxed{3.14}.")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Pi is \\boxed{3.14159}."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg, tolerance=0.01)
assert isinstance(result, EvaluateResult)
assert result.score == 1.0
# assert result.score == 1.0 # Corrected access
def test_wrong_answer_numeric(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="Answer: \\boxed{5}.")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Answer: \\boxed{4}."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score < 0.1
# assert result.score < 0.1 # Corrected access
def test_wrong_answer_mcq(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="Choice (B).")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Choice (A)."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
# The reason should be about the original message (ground_truth) if it's unparseable
assert (
result.reason is not None
and "Could not extract answers from original message (ground truth)" in result.reason
)
def test_no_answer_in_generated(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="Answer is (A).")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Answer is \\boxed{1}."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
# assert result.score == 0.0 # Duplicate line removed
assert result.reason is not None and "Could not extract answers from generated message" in result.reason
def test_no_answer_in_original_when_gen_is_unparseable(
self,
): # Renamed to avoid conflict and be more descriptive
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="I don't know.") # Gen unparseable
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Answer is \\boxed{1}." # GT parseable
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
assert result.reason is not None and "Could not extract answers from generated message" in result.reason
def test_unparseable_ground_truth(
self,
): # Renamed from the second test_no_answer_in_original
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="Answer is \\boxed{1}.") # Gen parseable
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "What is it?" # GT unparseable
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
# assert result.score == 0.0 # Duplicate line removed
assert (
result.reason is not None
and "Could not extract answers from original message (ground truth)" in result.reason
)
# --- Strictness Penalty Tests ---
def test_penalty_unboxed_or_issue1(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="The answer is $1/2 \\text{ or } 1$.")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "The answer is $\\boxed{1/2}$."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
# assert result.score == 0.0 # Duplicate line removed
assert result.reason is not None and result.reason.startswith("Strictness fail (Issue #1")
assert (
result.metrics["strictness_penalty_unboxed_or"].reason
== "Generated answer offers multiple numeric alternatives with an unboxed 'or' in the raw response."
)
def test_no_penalty_if_gen_is_single_boxed_or_expr(self):
user_messages = self._get_prompt() or []
assistant_message = Message(role="assistant", content="The answer is $\\boxed{1/2 \\text{ or } 1}$.")
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "The answer is $\\boxed{1/2 \\text{ or } 1}$."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 1.0
# assert result.score == 1.0 # Duplicate line removed
assert result.reason is None or ("Strictness fail" not in result.reason)
def test_penalty_ambiguity_issue2(self):
user_messages = self._get_prompt() or []
assistant_message = Message(
role="assistant",
content="The equation is $x=\\frac{1}{4}$. Some other numbers are 1, 2, 3.",
)
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "The answer is $\\boxed{1/4}$."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
# assert result.score == 0.0 # Duplicate line removed
assert result.reason is not None and result.reason.startswith("Strictness fail (Issue #2")
assert (
result.metrics["strictness_penalty_ambiguity"].reason
== "Ground truth is specific (one answer), but generated answer is ambiguous (multiple answers extracted, even after potential leniency)."
)
def test_issue_false_mcq_match_on_v_B(self):
user_content_str = "## Task B-1.3.\n\nA ship traveling along a river has covered $24 \\mathrm{~km}$ upstream and $28 \\mathrm{~km}$ downstream. For this journey, it took half an hour less than for traveling $30 \\mathrm{~km}$ upstream and $21 \\mathrm{~km}$ downstream, or half an hour more than for traveling $15 \\mathrm{~km}$ upstream and $42 \\mathrm{~km}$ downstream, assuming that both the ship and the river move uniformly.\n\nDetermine the speed of the ship in still water and the speed of the river."
assistant_content_generated_str = "## Solution.\n\nLet $t$ be the time required for the boat to travel $24 \\mathrm{~km}$ upstream and $28 \\mathrm{~km}$ downstream, $v_{R}$ the speed of the river, and $v_{B}$ the speed of the boat. When the boat is traveling upstream, its speed is $v_{B}-v_{R}$, and when it is traveling downstream, its speed is $v_{B}+v_{R}$.\n\nSince $t=\\frac{s}{v}$, from the given data, we obtain the following system of equations:\n\n$\\left\\{\\begin{array}{l}t=\\frac{24}{v_{B}-v_{R}}+\\frac{28}{v_{B}+v_{R}} \\\\ t+0.5=\\frac{30}{v_{B}-v_{R}}+\\frac{21}{v_{B}+v_{R}} \\\\ t-0.5=\\frac{15}{v_{B}-v_{R}}+\\frac{42}{v_{B}+v_{R}}\\end{array}\\right.$\n\nBy introducing new variables $x=\\frac{3}{v_{B}-v_{R}}, y=\\frac{7}{v_{B}+v_{R}}$, the system transforms into:\n\n$\\left\\{\\begin{array}{l}t=8 x+4 y \\\\ t+0.5=10 x+3 y \\\\ t-0.5=5 x+6 y\\end{array}\\right.$\n\nSubstituting $t$ from the first equation into the remaining two, we get:\n\n$\\left\\{\\begin{array}{l}8 x+4 y+0.5=10 x+3 y \\\\ 8 x+4 y-0.5=5 x+6 y\\end{array}\\right.$\n\n$\\left\\{\\begin{array}{l}2 x-y=0.5 \\\\ 3 x-2 y=0.5\\end{array}\\right.$\n\nThe solution to the last system is (0.5, 0.5). Then we have:\n\n$\\frac{3}{v_{B}-v_{R}}=0.5$, hence, $v_{B}-v_{R}=6 \\mathrm{~and}$\n\n$\\frac{7}{v_{B}+v_{R}}=0.5$, hence, $v_{B}+v_{R}=14$.\n\nThe speed of the river is $v_{R}=4 \\mathrm{~km} / \\mathrm{h}$, and the speed of the boat is $v_{B}=10 \\mathrm{~km} / \\mathrm{h}$.\n\n## Note:\n\nBy substituting $x=\\frac{1}{v_{B}-v_{R}}, y=\\frac{1}{v_{B}+v_{R}} \\mathrm{~and}$ following the same procedure, the initial system transforms into the system $\\left\\{\\begin{array}{l}6 x-7 y=0.5 \\\\ 9 x-14 y=0.5\\end{array}\\right.$\n\nThe solution to this system is $\\left(\\frac{1}{6}, \\frac{1}{14}\\right)$."
ground_truth_arg = (
"The speed of the river is $v_R=4 \\mathrm{km/h}$, and the speed of the boat is $v_B=10 \\mathrm{km/h}$."
)
user_message = Message(role="user", content=user_content_str)
assistant_message = Message(role="assistant", content=assistant_content_generated_str)
messages_arg = [user_message, assistant_message]
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
extracted_gen_reason = result.metrics["extracted_generated_answers"].reason
extracted_orig_reason = result.metrics["extracted_original_answers"].reason
assert "'B'" not in extracted_gen_reason
assert "'B'" not in extracted_orig_reason
assert result.score == 0.0
assert result.reason is not None and "Strictness fail (Conflicting Answers)" in result.reason
assert result.reason is not None and (
("0.5" in result.reason) or ("6.0" in result.reason) or ("14.0" in result.reason)
)
def test_no_penalty_ambiguity_if_gt_is_also_ambiguous(self):
user_messages = self._get_prompt() or []
assistant_message = Message(
role="assistant",
content="Solutions: $\\boxed{1}$, $\\boxed{2}$, $\\boxed{3}$.",
)
messages_arg = user_messages + [assistant_message]
ground_truth_arg = "Answers: $\\boxed{1}$, $\\boxed{2}$."
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
assert result.reason is not None and "Strictness fail (Conflicting Answers)" in result.reason
assert (
result.reason is not None
and "includes other distinct numerical values not matching any original answer: [3.0]" in result.reason
)
def test_issue3_scenario_correct_handling(self):
user_messages = self._get_prompt() or []
ground_truth_arg = (
"The speed of the river is $v_R=4 \\mathrm{km/h}$, and the speed of the boat is $v_B=10 \\mathrm{km/h}$."
)
response_content_str = "Let $x=\\frac{3}{v_{B}-v_{R}}, y=\\frac{7}{v_{B}+v_{R}}$. The solution to the last system is (0.5, 0.5). Then $v_R=4 \\mathrm{km} / \\mathrm{h}$, and $v_B=10 \\mathrm{km} / \\mathrm{h}$."
assistant_message = Message(role="assistant", content=response_content_str)
messages_arg = user_messages + [assistant_message]
result = math_reward(messages=messages_arg, ground_truth=ground_truth_arg)
assert isinstance(result, EvaluateResult)
assert result.score == 0.0
assert result.reason is not None and "Strictness fail (Conflicting Answers)" in result.reason
assert result.reason is not None and (
("0.5" in result.reason) or ("3.0" in result.reason) or ("7.0" in result.reason)
)
def test_require_units_basic_functionality(self):
user_messages = self._get_prompt() or []
ground_truth_arg = "Answer is \\boxed{10 km}."
assistant_message_match_units = Message(role="assistant", content="Answer is \\boxed{10 km}.")
messages_match_units_arg = user_messages + [assistant_message_match_units]
assistant_message_no_units = Message(role="assistant", content="Answer is \\boxed{10}.")
messages_no_units_arg = user_messages + [assistant_message_no_units]
result_match = math_reward(
messages=messages_match_units_arg,
ground_truth=ground_truth_arg,
require_units=True,
)
assert isinstance(result_match, EvaluateResult)
assert result_match.score == 1.0
assert result_match.reason is None or ("Unit presence mismatch" not in result_match.reason)
result_no_unit = math_reward(
messages=messages_no_units_arg,
ground_truth=ground_truth_arg,
require_units=True,
)
assert isinstance(result_no_unit, EvaluateResult)
assert result_no_unit.score == 0.0
assert result_no_unit.reason is not None and "Unit presence mismatch" in result_no_unit.reason