import unittest
from mindie_llm.runtime.config.huggingface_config import GenerationConfig, HuggingFaceConfig, RopeScaling
class TestGenerationConfig(unittest.TestCase):
def test_generation_config_from_dict_creates_instance(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 1,
"eos_token_id": 2
})
self.assertIsInstance(config, GenerationConfig)
self.assertEqual(config.pad_token_id, 0)
self.assertEqual(config.bos_token_id, 1)
self.assertEqual(config.eos_token_id, 2)
def test_generation_config_valid_token_ids(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 1,
"eos_token_id": 2
})
config.validate(vocab_size=10)
def test_generation_config_pad_token_id_negative_one_valid(self):
config = GenerationConfig.from_dict({
"pad_token_id": -1,
"bos_token_id": 1,
"eos_token_id": 2
})
config.validate(vocab_size=10)
def test_generation_config_bos_token_id_zero_valid(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 0,
"eos_token_id": 2
})
config.validate(vocab_size=10)
def test_generation_config_bos_token_id_out_of_bounds_raises(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 10,
"eos_token_id": 2
})
with self.assertRaises(ValueError):
config.validate(vocab_size=10)
def test_generation_config_eos_token_id_int_valid(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 1,
"eos_token_id": 2
})
config.validate(vocab_size=10)
def test_generation_config_eos_token_id_list_valid(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 1,
"eos_token_id": [2, 3, 4]
})
config.validate(vocab_size=10)
def test_generation_config_eos_token_id_nested_list_valid(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 1,
"eos_token_id": [[2, 3], [4, 5]]
})
config.validate(vocab_size=10)
def test_generation_config_eos_token_id_invalid_value_raises(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 1,
"eos_token_id": [10]
})
with self.assertRaises(ValueError):
config.validate(vocab_size=10)
def test_generation_config_validate_non_int_vocab_size_raises(self):
config = GenerationConfig.from_dict({
"pad_token_id": 0,
"bos_token_id": 1,
"eos_token_id": 2
})
with self.assertRaises(ValueError):
config.validate(vocab_size="10")
class TestHuggingFaceConfig(unittest.TestCase):
def test_from_dict_creates_instance(self):
config = HuggingFaceConfig.from_dict({
"max_position_embeddings": 2048,
"vocab_size": 32000,
})
self.assertIsInstance(config, HuggingFaceConfig)
self.assertEqual(config.max_position_embeddings, 2048)
self.assertEqual(config.vocab_size, 32000)
def test_valid_rope_scaling_passes_validation(self):
config = HuggingFaceConfig.from_dict({
"max_position_embeddings": 2048,
"vocab_size": 32000,
"rope_scaling": {
"factor": 4.0,
"rope_type": "yarn",
"original_max_position_embeddings": 4096,
"beta_fast": 64,
"beta_slow": 2
}
})
self.assertIsInstance(config.rope_scaling, RopeScaling)
def test_invalid_rope_type_raises(self):
with self.assertRaises(ValueError):
HuggingFaceConfig.from_dict({
"num_attention_heads": 32,
"rms_norm_eps": 0.1,
"max_position_embeddings": 2048,
"vocab_size": 32000,
"rope_scaling": {"rope_type": "invalid_type"}
})
def test_invalid_factor_out_of_bounds_raises(self):
with self.assertRaises(ValueError):
HuggingFaceConfig.from_dict({
"max_position_embeddings": 2048,
"vocab_size": 32000,
"rope_scaling": {"factor": 1e6}
})
def test_invalid_max_position_embeddings_negative_raises(self):
with self.assertRaises(ValueError):
HuggingFaceConfig.from_dict({
"num_attention_heads": 32,
"rms_norm_eps": 0.1,
"vocab_size": 32000,
"max_position_embeddings": -10
})
def test_invalid_vocab_size_zero_raises(self):
with self.assertRaises(ValueError):
HuggingFaceConfig.from_dict({
"num_attention_heads": 32,
"rms_norm_eps": 0.1,
"vocab_size": 0,
"max_position_embeddings": 2048
})
def test_pad_token_id_can_be_negative_one(self):
config = HuggingFaceConfig.from_dict({
"vocab_size": 32000,
"max_position_embeddings": 2048,
"pad_token_id": -1
})
self.assertEqual(config.pad_token_id, -1)
def test_eos_token_id_can_be_list_of_ints(self):
config = HuggingFaceConfig.from_dict({
"vocab_size": 32000,
"max_position_embeddings": 2048,
"eos_token_id": [1, 2, 3]
})
self.assertEqual(config.eos_token_id, [1, 2, 3])
def test_eos_token_id_can_be_nested_list(self):
config = HuggingFaceConfig.from_dict({
"vocab_size": 32000,
"max_position_embeddings": 2048,
"eos_token_id": [[1, 2], [3, 4]]
})
self.assertEqual(config.eos_token_id, [[1, 2], [3, 4]])
def test_none_rope_scaling_allowed(self):
config = HuggingFaceConfig.from_dict({
"vocab_size": 2800,
"max_position_embeddings": 2048,
"rope_scaling": None
})
self.assertIsInstance(config.rope_scaling, RopeScaling)
self.assertEqual(config.rope_scaling.factor, 1.0)
self.assertEqual(config.rope_scaling.rope_type, "linear")
self.assertIsNone(config.rope_scaling.original_max_position_embeddings)
self.assertEqual(config.rope_scaling.beta_fast, 32)
self.assertEqual(config.rope_scaling.beta_slow, 1)
def test_rope_scaling_default_values(self):
config = HuggingFaceConfig.from_dict({
"vocab_size": 2800,
"max_position_embeddings": 2048,
"rope_scaling": {}
})
self.assertIsInstance(config.rope_scaling, RopeScaling)
self.assertEqual(config.rope_scaling.factor, 1.0)
self.assertEqual(config.rope_scaling.rope_type, "linear")
self.assertIsNone(config.rope_scaling.original_max_position_embeddings)
self.assertEqual(config.rope_scaling.beta_fast, 32)
self.assertEqual(config.rope_scaling.beta_slow, 1)
if __name__ == '__main__':
unittest.main()
