add test transformer

src/rydberggpt/tests/test_transformer.py

@@ -1,49 +1,62 @@
-# import pytest
-# import torch
-# from torch import nn
+import pytest
 
-# from rydberggpt.models.transformer_wavefunction import TransformerWavefunction
-# from rydberggpt.training.loss import LabelSmoothing
+from rydberggpt.data.dataclasses import Batch
+from rydberggpt.data.loading import (
+    get_chunked_dataloader,
+    get_chunked_random_dataloader,
+    get_rydberg_dataloader,
+    get_streaming_dataloader,
+)
+from rydberggpt.models.rydberg_encoder_decoder import get_rydberg_graph_encoder_decoder
+from rydberggpt.training.trainer import RydbergGPTTrainer
+from rydberggpt.utils import create_config_from_yaml, load_yaml_file
 
 
-# def get_dummy_data():
-#     num_atoms = 3  # number of atoms
-#     num_samples = 8  # number of samples
+# Define a fixture for common model parameters (if needed)
+@pytest.fixture(scope="module")
+def config():
+    return {
+        "config_path": "configs/",
+    }
 
-#     H = torch.rand(
-#         (num_samples, num_atoms, 4), dtype=torch.float
-#     )  # [batch_size , num_atoms, 4]
-#     dataset = torch.randint(0, 2, (num_samples, num_atoms), dtype=torch.int64)
-#     return H, dataset
 
+# Define a fixture for common parameters
+@pytest.fixture(scope="module")
+def config_dataloader():
+    return {"data_path": "src/rydberggpt/tests/dataset_test/", "batch_size": 10}
 
-# def test_model_minimizes_loss():
-#     # prepare dummy data
-#     H, dataset = get_dummy_data()
-#     inputs = nn.functional.one_hot(dataset, 2)
-#     inputs = inputs.to(torch.float)
 
-#     # initialize model, criterion, and optimizer
-#     model = TransformerWavefunction(10, 2, 2)
-#     criterion = LabelSmoothing()
-#     optimizer = torch.optim.Adam(model.parameters(), lr=0.05)
+def test_rydberg_gpt_with_dataloader(config_dataloader):
+    yaml_dict = load_yaml_file("config/", "config_small")
+    config = create_config_from_yaml(yaml_dict)
+    dataloader, _ = get_chunked_random_dataloader(**config_dataloader)
 
-#     # calculate initial loss
-#     cond_probs = model.forward([H, inputs])
-#     first_loss = criterion(cond_probs, inputs)
+    # Create Model
+    model = get_rydberg_graph_encoder_decoder(config)
 
-#     # train the model for 50 iterations
-#     for _ in range(50):
-#         optimizer.zero_grad()
-#         cond_probs = model.forward([H, inputs])
-#         loss = criterion(cond_probs, inputs)
-#         loss.backward()
-#         optimizer.step()
+    # Initialize the trainer
+    rydberg_gpt_trainer = RydbergGPTTrainer(model, config)
 
-#     # check if the final loss is smaller than the initial loss
-#     assert loss < first_loss, "Final loss is not smaller than initial loss"
+    for i, batch in enumerate(dataloader):
+        assert isinstance(batch, Batch), "Batch is not an instance of the Batch class"
 
+        # Perform a forward pass through the model using the trainer
+        output = rydberg_gpt_trainer.training_step(batch, i)
 
-# if __name__ == "__main__":
-#     pytest.main([__file__])
-#     # test_model_minimizes_loss()
+        # Check output shapes, values, or any other property you are interested in
+        assert output >= 0, "Loss is not non-negative"
+
+        # Exit loop after one batch for quick testing
+        if i >= 1:
+            break
+
+
+if __name__ == "__main__":
+    pytest.main([__file__])
+
+    # config_dataloader = {
+    #     "data_path": "src/rydberggpt/tests/dataset_test/",
+    #     "batch_size": 10,
+    # }
+
+    # test_rydberg_gpt_with_dataloader(config_dataloader)