File size: 8,424 Bytes
2d7e335 | 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 | """Tests for AAM Diffusion Model components."""
import torch
import pytest
from diffusion_llm.config.model_config import AamDiffusionConfig, get_default_config, ModelConfig
from diffusion_llm.model.noise_scheduler import NoiseScheduler
from diffusion_llm.model.graph_encoder import GraphConditioningEncoder, GraphEncoderConfig
from diffusion_llm.model.diffusion_transformer import DiffusionTransformer
from diffusion_llm.model.aam_diffusion_model import AamDiffusionModel
from diffusion_llm.tokenizer.aam_tokenizer import AamTokenizer
class TestConfig:
"""Test configuration system."""
def test_default_config(self):
"""Test default configuration creation."""
config = get_default_config("base")
assert config.model.d_model == 768
assert config.model.n_layers == 12
assert config.diffusion.n_timesteps == 1000
def test_tiny_config(self):
"""Test tiny model configuration."""
config = get_default_config("tiny")
assert config.model.d_model == 256
assert config.model.n_layers == 4
def test_config_serialization(self, tmp_path):
"""Test config save/load roundtrip."""
config = get_default_config("small")
path = tmp_path / "config.json"
config.to_json(path)
loaded = AamDiffusionConfig.from_json(path)
assert loaded.model.d_model == config.model.d_model
assert loaded.model.n_layers == config.model.n_layers
def test_param_estimation(self):
"""Test parameter count estimation."""
config = ModelConfig(d_model=768, n_layers=12, d_ff=3072)
params = config.estimate_params()
assert "M" in params # Should be in millions
class TestTokenizer:
"""Test AAM Tokenizer."""
def test_basic_encoding(self):
"""Test basic text encoding."""
tokenizer = AamTokenizer()
# Train on sample text first
tokenizer.train(["Hello world this is a test", "Another test sentence"])
ids = tokenizer.encode("Hello world")
assert isinstance(ids, list)
assert len(ids) > 0
assert ids[0] == tokenizer.bos_id
assert ids[-1] == tokenizer.eos_id
def test_decode_roundtrip(self):
"""Test encode/decode roundtrip."""
tokenizer = AamTokenizer()
texts = [
"Berdasarkan analisis, pencuri adalah Diancang.",
"Anomali terdeteksi dalam laporan Hefei.",
"Evidence: Ju Jangmok, Snow Plum Pill.",
]
tokenizer.train(texts)
for text in texts:
ids = tokenizer.encode(text)
decoded = tokenizer.decode(ids, skip_special=True)
# Decoded text should contain key words
assert len(decoded) > 0
def test_special_tokens(self):
"""Test special token IDs."""
tokenizer = AamTokenizer()
assert tokenizer.pad_id == 0
assert tokenizer.bos_id == 1
assert tokenizer.eos_id == 2
def test_sentence_boundaries(self):
"""Test sentence boundary detection."""
tokenizer = AamTokenizer()
ids = [1, 10, 20, 5, 30, 40, 5, 50, 2] # BOS, sent, sent, EOS
boundaries = tokenizer.get_sentence_boundaries(ids)
assert 3 in boundaries # Index of <sent> token
assert 6 in boundaries
def test_save_load(self, tmp_path):
"""Test tokenizer save/load."""
tokenizer = AamTokenizer()
tokenizer.train(["Test text for tokenizer", "Another training example"])
path = tmp_path / "tokenizer.json"
tokenizer.save(path)
loaded = AamTokenizer.load(path)
assert loaded.vocab_size == tokenizer.vocab_size
assert loaded.is_trained
def test_structure_encoding(self):
"""Test encoding with graph structure tokens."""
tokenizer = AamTokenizer()
tokenizer.train(["Evidence text", "Anomaly description", "Reasoning step"])
ids = tokenizer.encode_with_structure(
text="Main narrative text",
evidence_nodes=["evidence1", "evidence2"],
anomalies=["anomaly1"],
)
assert isinstance(ids, list)
assert len(ids) > 0
def test_padding(self):
"""Test sequence padding."""
tokenizer = AamTokenizer()
ids = [1, 2, 3]
padded = tokenizer.pad_sequence(ids, max_len=10)
assert len(padded) == 10
assert padded[3:] == [0] * 7 # Padded with pad_id
class TestDiffusionTransformer:
"""Test Diffusion Transformer model."""
def test_forward_pass(self):
"""Test basic forward pass."""
config = ModelConfig(
d_model=128, n_layers=2, n_heads=4, d_ff=256,
vocab_size=1000, max_seq_len=64,
)
model = DiffusionTransformer(config)
x_t = torch.randn(2, 32, 128) # batch=2, seq=32, d=128
t = torch.tensor([100, 500])
output = model(x_t=x_t, t=t)
assert output.shape == (2, 32, 128)
def test_with_graph_conditioning(self):
"""Test forward pass with graph conditioning."""
config = ModelConfig(
d_model=128, n_layers=2, n_heads=4, d_ff=256,
vocab_size=1000, max_seq_len=64,
)
model = DiffusionTransformer(config)
x_t = torch.randn(2, 32, 128)
t = torch.tensor([100, 500])
graph_keys = torch.randn(2, 10, 128) # 10 graph nodes
graph_values = torch.randn(2, 10, 128)
output = model(x_t=x_t, t=t, graph_keys=graph_keys, graph_values=graph_values)
assert output.shape == (2, 32, 128)
class TestAamDiffusionModel:
"""Test complete AAM Diffusion Model."""
def test_model_creation_tiny(self):
"""Test creating a tiny model."""
config = get_default_config("tiny")
model = AamDiffusionModel(config)
n_params = model.get_num_params()
assert n_params > 0
assert n_params < 100e6 # Tiny should be under 100M
def test_forward_training(self):
"""Test training forward pass."""
config = get_default_config("tiny")
model = AamDiffusionModel(config)
model.eval()
token_ids = torch.randint(0, config.model.vocab_size, (2, 32))
timestep = torch.randint(0, config.diffusion.n_timesteps, (2,))
with torch.no_grad():
predicted, noise = model(token_ids=token_ids, timestep=timestep)
assert predicted.shape == noise.shape
def test_loss_computation(self):
"""Test loss computation."""
config = get_default_config("tiny")
model = AamDiffusionModel(config)
model.eval()
token_ids = torch.randint(0, config.model.vocab_size, (2, 32))
timestep = torch.randint(0, config.diffusion.n_timesteps, (2,))
with torch.no_grad():
predicted, noise = model(token_ids=token_ids, timestep=timestep)
loss = model.compute_loss(predicted, noise, timestep)
assert loss.item() >= 0
assert not torch.isnan(loss)
def test_save_load(self, tmp_path):
"""Test model save/load."""
config = get_default_config("tiny")
model = AamDiffusionModel(config)
path = str(tmp_path / "model.pt")
model.save(path)
loaded = AamDiffusionModel.load(path)
assert loaded.config.model.d_model == config.model.d_model
class TestGraphEncoder:
"""Test Graph Conditioning Encoder."""
def test_evidence_encoding(self):
"""Test encoding evidence nodes."""
config = GraphEncoderConfig(d_graph=128, n_graph_layers=2, n_graph_heads=4)
encoder = GraphConditioningEncoder(config, vocab_size=1000)
evidence_ids = torch.randint(0, 1000, (2, 5, 16)) # 2 batch, 5 nodes, 16 tokens each
evidence_conf = torch.tensor([[0.8, 0.6, 0.9, 0.7, 0.5],
[0.7, 0.8, 0.6, 0.9, 0.5]])
result = encoder(evidence_ids=evidence_ids, evidence_confidence=evidence_conf)
assert "keys" in result
assert "values" in result
def test_no_input(self):
"""Test encoder with no graph data (should return zeros)."""
config = GraphEncoderConfig(d_graph=128, n_graph_layers=2, n_graph_heads=4)
encoder = GraphConditioningEncoder(config, vocab_size=1000)
result = encoder()
assert "keys" in result
if __name__ == "__main__":
pytest.main([__file__, "-v"])
|