File size: 5,520 Bytes
5dd1bb4 | 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 | """Unit tests for training reward callables."""
from sql_env.training.rewards import (
reward_correctness,
reward_operational,
reward_progress,
)
def _completions(size: int) -> list[list[dict[str, str]]]:
return [[{"role": "assistant", "content": "QUERY: SELECT 1"}] for _ in range(size)]
def test_correctness_correct_answer() -> None:
result = reward_correctness(_completions(1), metadata=[{"correct": True}])
assert result == [1.0]
def test_correctness_wrong_answer() -> None:
result = reward_correctness(_completions(1), metadata=[{"correct": False}])
assert result == [0.0]
def test_correctness_no_answer() -> None:
result = reward_correctness(_completions(1), metadata=[{}])
assert result == [0.0]
def test_correctness_batch() -> None:
result = reward_correctness(
_completions(4),
metadata=[
{"answer_correct": True},
{"answer_correct": False},
{"correct": True},
{"correct": False},
],
)
assert result == [1.0, 0.0, 1.0, 0.0]
def test_correctness_empty_batch() -> None:
result = reward_correctness([])
assert result == []
def test_correctness_trl_compatible() -> None:
result = reward_correctness(_completions(2), metadata=[{"correct": True}, {}])
assert all(isinstance(item, float) for item in result)
def test_progress_full() -> None:
result = reward_progress(_completions(1), metadata=[{"progress": 1.0}])
assert result[0] == 1.0
def test_progress_none() -> None:
result = reward_progress(_completions(1), metadata=[{"progress": 0.0}])
assert result == [0.0]
def test_progress_partial() -> None:
result = reward_progress(_completions(1), metadata=[{"cumulative_progress": 0.4}])
assert 0.0 < result[0] < 1.0
def test_progress_normalized() -> None:
result = reward_progress(
_completions(4),
metadata=[
{"progress": -1.0},
{"progress": 0.2},
{"progress": 2.0},
{},
],
)
assert all(0.0 <= item <= 1.0 for item in result)
def test_progress_batch() -> None:
result = reward_progress(
_completions(3),
metadata=[{"progress": 0.0}, {"progress": 0.5}, {"progress": 1.0}],
)
assert result == [0.0, 0.5, 1.0]
def test_progress_trl_compatible() -> None:
result = reward_progress(_completions(2), metadata=[{}, {"progress": 0.1}])
assert all(isinstance(item, float) for item in result)
def test_operational_good_episode() -> None:
result = reward_operational(
_completions(1),
metadata=[
{
"operational_signals": [
{"exec_ok": True, "new_info": True, "repeat": False},
{"exec_ok": True, "new_info": False, "repeat": False},
]
}
],
)
assert result[0] > 0.0
def test_operational_all_errors() -> None:
result = reward_operational(
_completions(1),
metadata=[
{
"operational_signals": [
{"exec_ok": False, "new_info": False, "repeat": False},
{"exec_ok": False, "new_info": False, "repeat": False},
]
}
],
)
assert result[0] <= 0.0
def test_operational_repeat_penalty() -> None:
non_repeating = reward_operational(
_completions(1),
metadata=[
{
"operational_signals": [
{"exec_ok": True, "new_info": False, "repeat": False},
{"exec_ok": True, "new_info": False, "repeat": False},
]
}
],
)
repeating = reward_operational(
_completions(1),
metadata=[
{
"operational_signals": [
{"exec_ok": True, "new_info": False, "repeat": True},
{"exec_ok": True, "new_info": False, "repeat": True},
]
}
],
)
assert repeating[0] < non_repeating[0]
def test_operational_mixed_signals() -> None:
result = reward_operational(
_completions(1),
metadata=[
{
"operational_signals": [
{"exec_ok": True, "new_info": True, "repeat": False},
{"exec_ok": False, "new_info": False, "repeat": False},
{"exec_ok": True, "new_info": False, "repeat": True},
]
}
],
)
assert 0.0 < result[0] < 4.0
def test_operational_single_step() -> None:
result = reward_operational(
_completions(1),
metadata=[
{
"operational_signals": [
{"exec_ok": True, "new_info": False, "repeat": False}
]
}
],
)
assert isinstance(result[0], float)
def test_operational_batch() -> None:
result = reward_operational(
_completions(3),
metadata=[
{"operational": 1.0},
{"operational": -1.5},
{
"operational_signals": [
{"exec_ok": True, "new_info": True, "repeat": False},
]
},
],
)
assert len(result) == 3
assert result == [1.0, -1.5, 2.0]
def test_operational_trl_compatible() -> None:
result = reward_operational(_completions(2), metadata=[{}, {"operational": 0.5}])
assert all(isinstance(item, float) for item in result)
|