def test_AssertConfigIsValid_invalid_temperature_micros(abc_sampler_config):
"""Test that an error is thrown if temperature_micros is < 0."""
# Field not set.
abc_sampler_config.ClearField("temperature_micros")
with test.Raises(errors.UserError) as e_info:
samplers.Sampler(abc_sampler_config)
assert "Sampler.temperature_micros must be > 0" == str(e_info.value)
# Value is negative.
abc_sampler_config.temperature_micros = -1
with test.Raises(errors.UserError) as e_info:
samplers.Sampler(abc_sampler_config)
assert "Sampler.temperature_micros must be > 0" == str(e_info.value)
def test_AssertConfigIsValid_no_start_text(clgen_cache_dir,
abc_sampler_config):
"""Test that an error is thrown if start_text field is not set."""
del clgen_cache_dir
# Field not set.
abc_sampler_config.ClearField("start_text")
with test.Raises(errors.UserError) as e_info:
samplers.Sampler(abc_sampler_config)
assert "Sampler.start_text must be a string" == str(e_info.value)
# Value is an empty string.
abc_sampler_config.start_text = ""
with test.Raises(errors.UserError) as e_info:
samplers.Sampler(abc_sampler_config)
assert "Sampler.start_text must be a string" == str(e_info.value)
def __init__(self, config: clgen_pb2.Instance):
"""Instantiate an instance.
Args:
config: An Instance proto.
Raises:
UserError: If the instance proto contains invalid values, is missing
a model or sampler fields.
"""
try:
pbutil.AssertFieldIsSet(config, 'pretrained_model')
pbutil.AssertFieldIsSet(config, 'sampler')
except pbutil.ProtoValueError as e:
raise errors.UserError(e)
self.working_dir = None
if config.HasField('working_dir'):
self.working_dir: pathlib.Path = pathlib.Path(
os.path.expandvars(
config.working_dir)).expanduser().absolute()
# Enter a session so that the cache paths are set relative to any requested
# working directory.
with self.Session():
self.model: pretrained.PreTrainedModel = pretrained.PreTrainedModel(
pathlib.Path(config.pretrained_model))
self.sampler: samplers.Sampler = samplers.Sampler(config.sampler)
def __init__(self, config: clgen_pb2.Instance, dashboard_opts={}):
"""Instantiate an instance.
Args:
config: An Instance proto.
Raises:
UserError: If the instance proto contains invalid values, is missing
a model or sampler fields.
"""
try:
pbutil.AssertFieldIsSet(config, "model_specification")
pbutil.AssertFieldIsSet(config, "sampler")
except pbutil.ProtoValueError as e:
raise errors.UserError(e)
self.config = config
self.working_dir = None
if config.HasField("working_dir"):
self.working_dir: pathlib.Path = pathlib.Path(
os.path.expandvars(
config.working_dir)).expanduser().absolute()
# Enter a session so that the cache paths are set relative to any requested
# working directory.
with self.Session():
if config.HasField("model"):
self.model: models.Model = models.Model(config.model)
else:
self.model: pretrained.PreTrainedModel = pretrained.PreTrainedModel(
pathlib.Path(config.pretrained_model))
self.sampler: samplers.Sampler = samplers.Sampler(config.sampler)
self.dashboard = dashboard.Launch(**dashboard_opts)
def test_AssertConfigIsValid_invalid_batch_size(abc_sampler_config):
"""Test that an error is thrown if batch_size is < 1."""
# Field not set.
abc_sampler_config.ClearField("batch_size")
with test.Raises(errors.UserError) as e_info:
samplers.Sampler(abc_sampler_config)
assert "Sampler.batch_size must be > 0" == str(e_info.value)
# Value is zero.
abc_sampler_config.batch_size = 0
with test.Raises(errors.UserError) as e_info:
samplers.Sampler(abc_sampler_config)
assert "Sampler.batch_size must be > 0" == str(e_info.value)
# Value is negative.
abc_sampler_config.batch_size = -1
with test.Raises(errors.UserError) as e_info:
samplers.Sampler(abc_sampler_config)
assert "Sampler.batch_size must be > 0" == str(e_info.value)
def Train(self, *args, **kwargs) -> None:
with self.Session():
test_sampler_config = sampler_pb2.Sampler()
test_sampler_config.CopyFrom(self.sampler.config)
# Make all test samples the same 512-token length.
del test_sampler_config.termination_criteria[:]
test_sampler_config.termination_criteria.extend([
sampler_pb2.SampleTerminationCriterion(
maxlen=sampler_pb2.MaxTokenLength(
maximum_tokens_in_sample=512)),
])
test_sampler = samplers.Sampler(test_sampler_config)
# We inject the `test_sampler` argument so that we can create samples
# during training.
self.model.Train(*args, test_sampler=test_sampler, **kwargs)
def main(argv: typing.List[str]):
"""Main entry point."""
if len(argv) > 1:
raise app.UsageError("Unknown arguments: '{}'.".format(" ".join(
argv[1:])))
config = generative_model.CreateInstanceProtoFromFlags()
os.environ["CLGEN_CACHE"] = config.working_dir
logger = BacktrackingDatabaseLogger(backtracking_db.Database(FLAGS.db))
model = backtracking_model.BacktrackingModel(config.model, logger=logger)
sampler = samplers.Sampler(config.sampler)
sample_observers = generative_model.SampleObserversFromFlags()
model.Sample(sampler, sample_observers, seed=FLAGS.sample_seed)
def test_Sampler_Specialize_multiple_tokens_per(
abc_sampler_config: sampler_pb2.Sampler, ):
"""Test that InvalidSymtokTokens raised if depth tokens encode to mult."""
t = abc_sampler_config.termination_criteria.add()
t.symtok.depth_increase_token = "abc"
t.symtok.depth_decrease_token = "cba"
s = samplers.Sampler(abc_sampler_config)
def MockAtomizeString(string):
"""AtomizeString() with a multi-token output."""
del string
return np.array([1, 2, 3])
mock = AtomizerMock()
mock.AtomizeString = MockAtomizeString
with test.Raises(errors.InvalidSymtokTokens) as e_info:
s.Specialize(mock)
assert ("Sampler symmetrical depth tokens do not encode to a single "
"token using the corpus vocabulary")
def SampleStream():
config = generative_model.CreateInstanceProtoFromFlags()
os.environ["CLGEN_CACHE"] = config.working_dir
logger = MyLogger()
model = backtracking_model.BacktrackingModel(config.model, logger=logger)
model.corpus.Create()
sampler = samplers.Sampler(config.sampler)
atomizer = model.corpus.atomizer
sampler.Specialize(atomizer)
batch_size = model.backend.InitSampling(sampler, 0)
backtracker = backtracking_model.OpenClBacktrackingHelper(
atomizer, model._target_features)
model.backend.InitSampleBatch(sampler)
logger.OnSampleStart(backtracker)
yield from model.SampleOneWithBacktrackingToTextStream(
sampler, atomizer, backtracker)
def test_Sampler_Specialize_invalid_depth_tokens(
abc_sampler_config: sampler_pb2.Sampler, ):
"""Test that InvalidSymtokTokens raised if depth tokens cannot be encoded."""
t = abc_sampler_config.termination_criteria.add()
t.symtok.depth_increase_token = "{"
t.symtok.depth_decrease_token = "}"
s = samplers.Sampler(abc_sampler_config)
def MockAtomizeString(string):
"""AtomizeString() with a vocab error on depth tokens."""
if string == "{" or string == "}":
raise errors.VocabError()
else:
return np.ndarray([1])
mock = AtomizerMock()
mock.AtomizeString = MockAtomizeString
with test.Raises(errors.InvalidSymtokTokens) as e_info:
s.Specialize(mock)
assert ("Sampler symmetrical depth tokens cannot be encoded using the "
"corpus vocabulary") == str(e_info.value)
def test_Sampler_Specialize_encoded_start_text(
abc_sampler_config: sampler_pb2.Sampler, ):
s = samplers.Sampler(abc_sampler_config)
assert s.encoded_start_text is None
s.Specialize(AtomizerMock())
np.testing.assert_array_equal(np.array([1]), s.encoded_start_text)
def test_Sampler_batch_size(abc_sampler_config: sampler_pb2.Sampler):
"""Test that batch_size is set from Sampler proto."""
abc_sampler_config.batch_size = 99
s = samplers.Sampler(abc_sampler_config)
assert 99 == s.batch_size
def test_Sampler_temperature(abc_sampler_config: sampler_pb2.Sampler):
"""Test that temperature is set from Sampler proto."""
abc_sampler_config.temperature_micros = 1000000
s = samplers.Sampler(abc_sampler_config)
assert pytest.approx(1.0) == s.temperature
def test_Sampler_start_text(abc_sampler_config: sampler_pb2.Sampler):
"""Test that start_text is set from Sampler proto."""
s = samplers.Sampler(abc_sampler_config)
assert s.start_text == abc_sampler_config.start_text
def test_Sampler_config_type_error():
"""Test that a TypeError is raised if config is not a Sampler proto."""
with test.Raises(TypeError) as e_info:
samplers.Sampler(1)
assert "Config must be a Sampler proto. Received: 'int'" == str(
e_info.value)
