def test_computation_target(temp_output_dir, train_opts, target_opts, atol):
check_computation(
Estimator,
temp_output_dir,
train_opts,
target_opts,
output_name=constants.TARGET_TAGS,
expected_avg_probs=0.426000,
atol=atol,
)
# Testing resuming training
resume_opts = argparse.Namespace(**vars(train_opts))
resume_opts.save_model = False
resume_opts.checkpoint_save = True
resume_opts.resume = True
resume_opts.epochs += 1
check_computation(
Estimator,
temp_output_dir,
resume_opts,
target_opts,
output_name=constants.TARGET_TAGS,
expected_avg_probs=0.426000,
atol=atol,
)
def test_computation_target(
tmp_path,
xlm_model,
xlm_tokenizer,
xlm_model_dir,
xlm_config_dict,
train_config,
data_config,
big_atol,
):
train_config['run']['output_dir'] = tmp_path
train_config['data'] = data_config
train_config['system'] = xlm_config_dict
xlm_model.save_pretrained(tmp_path)
xlm_tokenizer.save_pretrained(tmp_path)
train_config['system']['model']['encoder']['model_name'] = str(tmp_path)
check_computation(
train_config,
tmp_path,
output_name=const.TARGET_TAGS,
expected_avg_probs=0.410072,
atol=big_atol,
)
def test_computation_target(
tmp_path,
xlmr_model,
xlmr_model_dir,
xlmr_config_dict,
train_config,
data_config,
big_atol,
):
train_config['run']['output_dir'] = tmp_path
train_config['data'] = data_config
train_config['system'] = xlmr_config_dict
xlmr_model.save_pretrained(tmp_path)
train_config['system']['model']['encoder']['model_name'] = str(tmp_path)
# When using `adamw` optimizer and the `optimizer.training_steps` are not set:
with pytest.raises(ValueError):
check_computation(
train_config,
tmp_path,
output_name=const.TARGET_TAGS,
expected_avg_probs=0.383413,
atol=big_atol,
)
# Now training will run:
train_config['system']['optimizer']['training_steps'] = 10
check_computation(
train_config,
tmp_path,
output_name=const.TARGET_TAGS,
expected_avg_probs=0.383413,
atol=big_atol,
)
def test_computation_source(temp_output_dir, train_opts, source_opts, atol):
check_computation(
Estimator,
temp_output_dir,
train_opts,
source_opts,
output_name=constants.SOURCE_TAGS,
expected_avg_probs=0.456631,
atol=atol,
)
def test_computation_gaps(temp_output_dir, train_opts, gap_opts, atol):
check_computation(
NuQE,
temp_output_dir,
train_opts,
gap_opts,
output_name=constants.GAP_TAGS,
expected_avg_probs=0.454558,
atol=atol,
)
def test_computation_target(temp_output_dir, train_opts, target_opts, atol):
check_computation(
NuQE,
temp_output_dir,
train_opts,
target_opts,
output_name=constants.TARGET_TAGS,
expected_avg_probs=0.466939,
atol=atol,
)
def test_computation(temp_output_dir, train_opts, nuqe_opts, atol):
check_computation(
NuQE,
temp_output_dir,
train_opts,
nuqe_opts,
output_name=constants.TARGET_TAGS,
expected_avg_probs=0.572441,
atol=atol,
)
def test_computation(temp_output_dir, train_opts, quetch_opts, atol):
check_computation(
QUETCH,
temp_output_dir,
train_opts,
quetch_opts,
output_name=constants.TARGET_TAGS,
expected_avg_probs=0.439731,
atol=atol,
)
def test_computation_source(temp_output_dir, train_opts, source_opts, atol):
check_computation(
QUETCH,
temp_output_dir,
train_opts,
source_opts,
output_name=constants.SOURCE_TAGS,
expected_avg_probs=0.355306,
atol=atol,
)
def test_computation_gaps(temp_output_dir, train_opts, gap_opts, atol):
check_computation(
QUETCH,
temp_output_dir,
train_opts,
gap_opts,
output_name=constants.GAP_TAGS,
expected_avg_probs=0.251563,
atol=atol,
)
def test_computation_targetgaps(tmp_path, output_targetgaps_config,
train_config, data_config, big_atol):
train_config['data'] = data_config
train_config['system'] = output_targetgaps_config
check_computation(
train_config,
tmp_path,
output_name=const.TARGET_TAGS,
expected_avg_probs=0.507699,
atol=big_atol,
)
def test_computation_source(tmp_path, output_source_config, train_config,
data_config, big_atol):
train_config['data'] = data_config
train_config['system'] = output_source_config
check_computation(
train_config,
tmp_path,
output_name=const.SOURCE_TAGS,
expected_avg_probs=0.486522,
atol=big_atol,
)
def test_computation_gaps(tmp_path, output_gaps_config, train_config,
data_config, atol):
train_config['data'] = data_config
train_config['system'] = output_gaps_config
check_computation(
train_config,
tmp_path,
output_name=const.GAP_TAGS,
expected_avg_probs=0.316064,
atol=atol,
)
def test_computation_target(tmp_path, output_target_config, train_config,
data_config, atol):
train_config['data'] = data_config
train_config['system'] = output_target_config
check_computation(
train_config,
tmp_path,
output_name=const.TARGET_TAGS,
expected_avg_probs=0.498354,
atol=atol,
)
def test_computation_gaps(temp_output_dir, train_opts, gap_opts, atol):
gap_opts.predict_target = False
check_computation(
Estimator,
temp_output_dir,
train_opts,
gap_opts,
output_name=constants.GAP_TAGS,
expected_avg_probs=0.322811,
atol=atol,
)
gap_opts.predict_target = True
check_computation(
Estimator,
temp_output_dir,
train_opts,
gap_opts,
output_name=constants.GAP_TAGS,
expected_avg_probs=0.328905,
atol=atol,
)
def test_computation_target(
tmp_path,
bert_model,
bert_model_dir,
bert_config_dict,
train_config,
data_config,
big_atol,
):
train_config['run']['output_dir'] = tmp_path
train_config['data'] = data_config
train_config['system'] = bert_config_dict
shutil.copy2(bert_model_dir / VOCAB_FILES_NAMES['vocab_file'], tmp_path)
bert_model.save_pretrained(tmp_path)
train_config['system']['model']['encoder']['model_name'] = str(tmp_path)
check_computation(
train_config,
tmp_path,
output_name=const.TARGET_TAGS,
expected_avg_probs=0.550805,
atol=big_atol,
)