Class LmInferenceTest

lm_inference_test.py:31–85

class LmInferenceTest(unittest.TestCase):

  @classmethod
  def setUpClass(cls):
    super().setUpClass()
    gin_file = [
        'base_htrans.gin',
        'size/medium_150M.gin',
        'options/positions_t5.gin',
        'options/lr_cosine_decay.gin',
        'options/seq_1024_nocache.gin',
        'geometry_150M_generate.gin',
    ]

    gin_param = [
        'DecoderOnlyLanguageModelGenerate.output_token_losses=True',
        'TransformerTaskConfig.batch_size=2',
        'TransformerTaskConfig.sequence_length=128',
        'Trainer.restore_state_variables=False',
    ]

    gin_search_paths = [
        os.path.join(_MELIAD_PATH.value, 'transformer/configs'),
        os.getcwd(),
    ]

    vocab_path = os.path.join(_DATA_PATH.value, 'geometry.757.model')

    lm.parse_gin_configuration(gin_file, gin_param, gin_paths=gin_search_paths)

    cls.loaded_lm = lm.LanguageModelInference(
        vocab_path, _DATA_PATH.value, mode='beam_search'
    )

  def test_lm_decode(self):
    outputs = LmInferenceTest.loaded_lm.beam_decode(
        '{S} a : ; b : ; c : ; d : T a b c d 00 T a c b d 01 ? T a d b c'
        ' {F1} x00',
        eos_tokens=[';'],
    )
    self.assertEqual(
        outputs['seqs_str'],
        ['e : D a b c e 02 D a c b e 03 ;', 'e : C a c e 02 C b d e 03 ;'],
    )

  def test_lm_score_may_fail_numerically_for_external_meliad(self):
    outputs = LmInferenceTest.loaded_lm.beam_decode(
        '{S} a : ; b : ; c : ; d : T a b c d 00 T a c b d 01 ? T a d b c'
        ' {F1} x00',
        eos_tokens=[';'],
    )
    self.assertEqual(
        outputs['scores'],
        [-1.18607294559478759765625, -1.10228693485260009765625],
    )


if __name__ == '__main__':