File size: 18,823 Bytes
1fa3c6c | 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 | # Copyright 2020-2026 The HuggingFace Team. All rights reserved.
#
# 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 tempfile
from dataclasses import dataclass
from unittest.mock import mock_open, patch
import pytest
from datasets import DatasetDict, load_dataset
from trl import DatasetMixtureConfig, TrlParser, get_dataset
from trl.scripts.utils import DatasetConfig
from .testing_utils import TrlTestCase
@dataclass
class MyDataclass:
arg1: int
arg2: str = "default"
@dataclass
class InvalidDataclass:
config: str # This should raise an error in the TrlParser
class TestTrlParser(TrlTestCase):
def test_init_without_config_field(self):
"""Test initialization without 'config' field in the dataclasses."""
parser = TrlParser(dataclass_types=[MyDataclass])
assert isinstance(parser, TrlParser)
def test_init_with_config_field(self):
"""Test initialization with a 'config' field in the dataclass (should raise ValueError)."""
with pytest.raises(ValueError, match="has a field named 'config'"):
TrlParser(dataclass_types=[InvalidDataclass])
@patch("builtins.open", mock_open(read_data="env:\n VAR1: value1\n VAR2: value2\narg1: 2"))
@patch("yaml.safe_load")
@patch("os.environ", new_callable=dict) # Mock os.environ as a dictionary
def test_parse_args_and_config_with_valid_config(self, mock_environ, mock_yaml_load):
"""Test parse_args_and_config method with valid arguments and config."""
mock_yaml_load.return_value = {"env": {"VAR1": "value1", "VAR2": "value2"}, "arg1": 2}
parser = TrlParser(dataclass_types=[MyDataclass])
args = ["--arg2", "value", "--config", "config.yaml"] # don't set arg1 to test default value
# Simulate the config being loaded and environment variables being set
result_args = parser.parse_args_and_config(args)
# Set the environment variables using the mock
mock_environ["VAR1"] = "value1"
mock_environ["VAR2"] = "value2"
# Ensure that the environment variables were set correctly
assert mock_environ.get("VAR1") == "value1"
assert mock_environ.get("VAR2") == "value2"
# Check the parsed arguments
assert len(result_args) == 1
assert isinstance(result_args[0], MyDataclass)
assert result_args[0].arg1 == 2
assert result_args[0].arg2 == "value"
@patch("builtins.open", mock_open(read_data="arg1: 2"))
@patch("yaml.safe_load")
def test_parse_args_and_arg_override_config(self, mock_yaml_load):
"""Test parse_args_and_config method and check that arguments override the config."""
mock_yaml_load.return_value = {"arg1": 2} # this arg is meant to be overridden
parser = TrlParser(dataclass_types=[MyDataclass])
args = ["--arg1", "3", "--config", "config.yaml"] # override arg1 default with 3
# Simulate the config being loaded and arguments being passed
result_args = parser.parse_args_and_config(args)
# Check the parsed arguments
assert len(result_args) == 1
assert isinstance(result_args[0], MyDataclass)
assert result_args[0].arg1 == 3
@patch("builtins.open", mock_open(read_data="env: not_a_dict"))
@patch("yaml.safe_load")
def test_parse_args_and_config_with_invalid_env(self, mock_yaml_load):
"""Test parse_args_and_config method when the 'env' field is not a dictionary."""
mock_yaml_load.return_value = {"env": "not_a_dict"}
parser = TrlParser(dataclass_types=[MyDataclass])
args = ["--arg1", "2", "--arg2", "value", "--config", "config.yaml"]
with pytest.raises(ValueError, match="`env` field should be a dict in the YAML file."):
parser.parse_args_and_config(args)
def test_parse_args_and_config_without_config(self):
"""Test parse_args_and_config without the `--config` argument."""
parser = TrlParser(dataclass_types=[MyDataclass])
args = ["--arg1", "2", "--arg2", "value"]
# Simulate no config, just parse args normally
result_args = parser.parse_args_and_config(args)
# Check that the arguments are parsed as is
assert len(result_args) == 1
assert isinstance(result_args[0], MyDataclass)
assert result_args[0].arg1 == 2
assert result_args[0].arg2 == "value"
def test_set_defaults_with_config(self):
"""Test set_defaults_with_config updates the defaults."""
parser = TrlParser(dataclass_types=[MyDataclass])
# Update defaults
parser.set_defaults_with_config(arg1=42)
# Ensure the default value is updated
result_args = parser.parse_args_and_config([])
assert len(result_args) == 1
assert isinstance(result_args[0], MyDataclass)
assert result_args[0].arg1 == 42
def test_parse_args_and_config_with_remaining_strings(self):
parser = TrlParser(dataclass_types=[MyDataclass])
args = ["--arg1", "2", "--arg2", "value", "remaining"]
# Simulate no config, just parse args normally
result_args = parser.parse_args_and_config(args, return_remaining_strings=True)
# Check that the arguments are parsed as is
assert len(result_args) == 2
assert isinstance(result_args[0], MyDataclass)
assert result_args[0].arg1 == 2
assert result_args[0].arg2 == "value"
assert result_args[1] == ["remaining"]
@patch("builtins.open", mock_open(read_data="remaining_string_in_config: abc"))
@patch("yaml.safe_load")
def test_parse_args_and_config_with_remaining_strings_in_config_and_args(self, mock_yaml_load):
mock_yaml_load.return_value = {"remaining_string_in_config": "abc"}
parser = TrlParser(dataclass_types=[MyDataclass])
args = ["--arg1", "2", "--remaining_string_in_args", "def", "--config", "config.yaml"]
# Simulate the config being loaded and arguments being passed
result_args = parser.parse_args_and_config(args, return_remaining_strings=True)
# Check that the arguments are parsed as is
assert len(result_args) == 2
assert isinstance(result_args[0], MyDataclass)
assert result_args[0].arg1 == 2
assert result_args[1] == ["--remaining_string_in_config", "abc", "--remaining_string_in_args", "def"]
@patch("builtins.open", mock_open(read_data="arg1: 2\narg2: config_value"))
@patch("yaml.safe_load")
def test_subparsers_with_config_defaults(self, mock_yaml_load):
"""Test that config defaults are applied to all subparsers."""
mock_yaml_load.return_value = {"arg1": 2, "arg2": "config_value"}
# Create the main parser
parser = TrlParser()
# Add subparsers
subparsers = parser.add_subparsers(dest="command", parser_class=TrlParser)
# Create a subparser for a specific command
subparsers.add_parser("subcommand", dataclass_types=[MyDataclass])
# Parse with config file
args = ["subcommand", "--config", "config.yaml"]
result_args = parser.parse_args_and_config(args)
# Check main parser arguments
assert len(result_args) == 1
# Check that config values were applied to the subparser
assert result_args[0].arg1 == 2 # Default from config
assert result_args[0].arg2 == "config_value" # Default from config
@patch("builtins.open", mock_open(read_data="arg1: 2\narg2: config_value"))
@patch("yaml.safe_load")
def test_subparsers_with_config_defaults_and_arg_override(self, mock_yaml_load):
"""Test that config defaults are applied to all subparsers."""
mock_yaml_load.return_value = {"arg1": 2, "arg2": "config_value"}
# Create the main parser
parser = TrlParser()
# Add subparsers
subparsers = parser.add_subparsers(dest="command", parser_class=TrlParser)
# Create a subparser for a specific command
subparsers.add_parser("subcommand", dataclass_types=[MyDataclass])
# Test with command line arguments overriding config
args = ["subcommand", "--arg1", "3", "--config", "config.yaml"]
result_args = parser.parse_args_and_config(args)
# Command line arguments should override config
assert result_args[0].arg1 == 3
assert result_args[0].arg2 == "config_value" # Still from config
@patch("builtins.open", mock_open(read_data="arg1: 2\nthis_arg_does_not_exist: config_value"))
@patch("yaml.safe_load")
def test_subparsers_with_config_defaults_and_arg_override_wrong_name(self, mock_yaml_load):
"""Test that config defaults are applied to all subparsers."""
mock_yaml_load.return_value = {"arg1": 2, "this_arg_does_not_exist": "config_value"}
# Create the main parser
parser = TrlParser()
# Add subparsers
subparsers = parser.add_subparsers(dest="command", parser_class=TrlParser)
# Create a subparser for a specific command
subparsers.add_parser("subcommand", dataclass_types=[MyDataclass])
# Test with command line arguments overriding config
args = ["subcommand", "--arg1", "3", "--config", "config.yaml"]
with pytest.raises(ValueError):
parser.parse_args_and_config(args)
parser.parse_args_and_config(args, fail_with_unknown_args=False)
@patch("builtins.open", mock_open(read_data="arg1: 2\narg2: config_value"))
@patch("yaml.safe_load")
def test_subparsers_multiple_with_config_defaults(self, mock_yaml_load):
"""Test that config defaults are applied to all subparsers."""
mock_yaml_load.return_value = {"arg1": 2, "arg2": "config_value"}
# Create the main parser
parser = TrlParser()
# Add subparsers
subparsers = parser.add_subparsers(dest="command", parser_class=TrlParser)
# Create a subparser for a specific command
subparsers.add_parser("subcommand0", dataclass_types=[MyDataclass])
subparsers.add_parser("subcommand1", dataclass_types=[MyDataclass])
for idx in range(2):
# Parse with config file
args = [f"subcommand{idx}", "--config", "config.yaml"]
result_args = parser.parse_args_and_config(args)
# Check main parser arguments
assert len(result_args) == 1
# Check that config values were applied to the subparser
assert result_args[0].arg1 == 2 # Default from config
assert result_args[0].arg2 == "config_value" # Default from config
class TestGetDataset:
def test_single_dataset_with_config(self):
mixture_config = DatasetMixtureConfig(
datasets=[DatasetConfig(path="trl-internal-testing/zen", name="standard_language_modeling")]
)
result = get_dataset(mixture_config)
expected = load_dataset("trl-internal-testing/zen", "standard_language_modeling")
assert expected["train"][:] == result["train"][:]
def test_single_dataset_preference_config(self):
mixture_config = DatasetMixtureConfig(
datasets=[DatasetConfig(path="trl-internal-testing/zen", name="standard_preference")]
)
result = get_dataset(mixture_config)
expected = load_dataset("trl-internal-testing/zen", "standard_preference")
assert expected["train"][:] == result["train"][:]
def test_single_dataset_streaming(self):
mixture_config = DatasetMixtureConfig(
datasets=[DatasetConfig(path="trl-internal-testing/zen", name="standard_language_modeling")],
streaming=True,
)
result = get_dataset(mixture_config)
expected = load_dataset("trl-internal-testing/zen", "standard_language_modeling")
assert expected["train"].to_list() == list(result["train"])
def test_dataset_mixture_basic(self):
dataset_config1 = DatasetConfig(
path="trl-internal-testing/zen", name="standard_prompt_completion", split="train", columns=["prompt"]
)
dataset_config2 = DatasetConfig(
path="trl-internal-testing/zen", name="standard_preference", split="train", columns=["prompt"]
)
mixture_config = DatasetMixtureConfig(datasets=[dataset_config1, dataset_config2])
result = get_dataset(mixture_config)
assert isinstance(result, DatasetDict)
assert "train" in result
train_dataset = result["train"]
assert train_dataset.column_names == ["prompt"]
prompts = train_dataset["prompt"]
expected_first_half = load_dataset("trl-internal-testing/zen", "standard_preference", split="train")
assert prompts[: len(prompts) // 2] == expected_first_half["prompt"]
expected_second_half = load_dataset("trl-internal-testing/zen", "standard_prompt_completion", split="train")
assert prompts[len(prompts) // 2 :] == expected_second_half["prompt"]
def test_dataset_mixture_with_weights(self):
dataset_config1 = DatasetConfig(
path="trl-internal-testing/zen", name="standard_prompt_completion", split="train[:50%]", columns=["prompt"]
)
dataset_config2 = DatasetConfig(
path="trl-internal-testing/zen", name="standard_preference", split="train[:50%]", columns=["prompt"]
)
mixture_config = DatasetMixtureConfig(datasets=[dataset_config1, dataset_config2])
result = get_dataset(mixture_config)
assert isinstance(result, DatasetDict)
assert "train" in result
train_dataset = result["train"]
assert train_dataset.column_names == ["prompt"]
prompts = train_dataset["prompt"]
expected_first_half = load_dataset("trl-internal-testing/zen", "standard_preference", split="train[:50%]")
assert prompts[: len(prompts) // 2] == expected_first_half["prompt"]
expected_second_half = load_dataset(
"trl-internal-testing/zen", "standard_prompt_completion", split="train[:50%]"
)
assert prompts[len(prompts) // 2 :] == expected_second_half["prompt"]
def test_dataset_mixture_with_test_split(self):
mixture_config = DatasetMixtureConfig(
datasets=[DatasetConfig(path="trl-internal-testing/zen", name="standard_language_modeling")],
test_split_size=2,
)
result = get_dataset(mixture_config)
assert isinstance(result, DatasetDict)
assert "train" in result
assert "test" in result
assert len(result["train"]) == 15
assert len(result["test"]) == 2
def test_empty_dataset_mixture_raises_error(self):
mixture_config = DatasetMixtureConfig(datasets=[])
with pytest.raises(ValueError, match="No datasets were loaded"):
get_dataset(mixture_config)
def test_mixture_multiple_different_configs(self):
dataset_config1 = DatasetConfig(
path="trl-internal-testing/zen", name="conversational_preference", split="train", columns=["prompt"]
)
dataset_config2 = DatasetConfig(
path="trl-internal-testing/zen", name="conversational_prompt_only", split="test"
)
mixture_config = DatasetMixtureConfig(datasets=[dataset_config1, dataset_config2])
result = get_dataset(mixture_config)
assert isinstance(result, DatasetDict)
assert "train" in result
assert len(result["train"]) > 0
def test_trlparser_parses_yaml_config_correctly(self):
# Prepare YAML content exactly like your example
# docstyle-ignore
yaml_content = """
datasets:
- path: trl-internal-testing/zen
name: standard_prompt_only
- path: trl-internal-testing/zen
name: standard_preference
columns:
- prompt
"""
# Write YAML to a temporary file
with tempfile.NamedTemporaryFile("w+", suffix=".yaml") as tmpfile:
tmpfile.write(yaml_content)
tmpfile.flush()
parser = TrlParser((DatasetMixtureConfig,))
args = parser.parse_args_and_config(args=["--config", tmpfile.name])[0]
# Assert that we got DatasetMixtureConfig instance
assert isinstance(args, DatasetMixtureConfig)
# Assert datasets list length
assert len(args.datasets) == 2
# Check first dataset
dataset_config1 = args.datasets[0]
assert isinstance(dataset_config1, DatasetConfig)
assert dataset_config1.path == "trl-internal-testing/zen"
assert dataset_config1.name == "standard_prompt_only"
assert dataset_config1.columns is None # No columns specified
# Check second dataset
dataset_config2 = args.datasets[1]
assert isinstance(dataset_config2, DatasetConfig)
assert dataset_config2.path == "trl-internal-testing/zen"
assert dataset_config2.name == "standard_preference"
assert dataset_config2.columns == ["prompt"] # Columns specified
def test_trlparser_parses_yaml_and_loads_dataset(self):
# Prepare YAML content exactly like your example
# docstyle-ignore
yaml_content = """
datasets:
- path: trl-internal-testing/zen
name: standard_language_modeling
"""
# Write YAML to a temporary file
with tempfile.NamedTemporaryFile("w+", suffix=".yaml") as tmpfile:
tmpfile.write(yaml_content)
tmpfile.flush()
parser = TrlParser((DatasetMixtureConfig,))
args = parser.parse_args_and_config(args=["--config", tmpfile.name])[0]
# Load the dataset using get_dataset
result = get_dataset(args)
expected = load_dataset("trl-internal-testing/zen", "standard_language_modeling")
assert expected["train"][:] == result["train"][:]
|