diffusion-model-framework 1.1.0rc1__tar.gz → 1.1.0rc2__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (58) hide show
  1. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/.gitignore +7 -0
  2. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/PKG-INFO +1 -1
  3. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/assets/sampling-small.gif +1 -1
  4. diffusion_model_framework-1.1.0rc2/diffusion_models/__init__.py +1 -0
  5. diffusion_model_framework-1.1.0rc2/diffusion_models/_store/__init__.py +15 -0
  6. diffusion_model_framework-1.1.0rc2/diffusion_models/_store/beta_scheduler.py +15 -0
  7. diffusion_model_framework-1.1.0rc2/diffusion_models/_store/dataset.py +40 -0
  8. diffusion_model_framework-1.1.0rc2/diffusion_models/_store/diffuser.py +20 -0
  9. diffusion_model_framework-1.1.0rc2/diffusion_models/_store/model.py +14 -0
  10. diffusion_model_framework-1.1.0rc2/diffusion_models/_store/optimizer.py +30 -0
  11. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/diffusion_trainer.py +48 -8
  12. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/gaussian_diffusion/ddimm_diffuser.py +1 -1
  13. diffusion_model_framework-1.1.0rc2/diffusion_models/utils/adopt.py +501 -0
  14. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/utils/schemas.py +4 -0
  15. diffusion_model_framework-1.1.0rc2/diffusion_models/utils/tensorboard.py +78 -0
  16. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/source/conf.py +1 -1
  17. diffusion_model_framework-1.1.0rc2/examples/config/baseline.yaml +40 -0
  18. diffusion_model_framework-1.1.0rc2/examples/train_model.py +118 -0
  19. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/pyproject.toml +5 -1
  20. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/uv.lock +56 -1
  21. diffusion_model_framework-1.1.0rc1/diffusion_models/__init__.py +0 -1
  22. diffusion_model_framework-1.1.0rc1/diffusion_models/utils/adopt.py +0 -501
  23. diffusion_model_framework-1.1.0rc1/diffusion_models/utils/tensorboard.py +0 -45
  24. diffusion_model_framework-1.1.0rc1/examples/train_model.py +0 -118
  25. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/.github/workflows/build-and-release-docs.yaml +0 -0
  26. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/.github/workflows/release.yaml +0 -0
  27. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/.github/workflows/run-tests.yaml +0 -0
  28. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/CHANGELOG.md +0 -0
  29. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/README.md +0 -0
  30. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/diffusion_inference.py +0 -0
  31. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/gaussian_diffusion/__init__.py +0 -0
  32. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/gaussian_diffusion/base_diffuser.py +0 -0
  33. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/gaussian_diffusion/beta_schedulers.py +0 -0
  34. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/gaussian_diffusion/gaussian_diffuser.py +0 -0
  35. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/models/SimpleUnet.py +0 -0
  36. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/models/__init__.py +0 -0
  37. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/models/base_diffusion_model.py +0 -0
  38. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/models/conditioned_simple_unet.py +0 -0
  39. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/utils/__init__.py +0 -0
  40. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/diffusion_models/utils/batch_processor.py +0 -0
  41. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/Makefile +0 -0
  42. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/make.bat +0 -0
  43. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/source/_templates/versions.html +0 -0
  44. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/source/custom/post-modules/.gitkeep +0 -0
  45. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/source/custom/pre-modules/introduction.md +0 -0
  46. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/source/examples.rst +0 -0
  47. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/docs/source/index.rst +0 -0
  48. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/examples/__init__.py +0 -0
  49. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/examples/coco_playground.py +0 -0
  50. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/examples/infer.py +0 -0
  51. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/__init__.py +0 -0
  52. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/conftest.py +0 -0
  53. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/gaussian_diffusion/__init__.py +0 -0
  54. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/gaussian_diffusion/test_beta_schedulers.py +0 -0
  55. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/test_base.py +0 -0
  56. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/utils/__init__.py +0 -0
  57. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/utils/test_schemas.py +0 -0
  58. {diffusion_model_framework-1.1.0rc1 → diffusion_model_framework-1.1.0rc2}/tests/utils/test_tensorboard.py +0 -0
@@ -165,3 +165,10 @@ test_results/
165
165
  data/
166
166
  checkpoints/
167
167
  runs/
168
+
169
+ # Local artifacts: model weights, generated media and scratch notebooks
170
+ save/
171
+ *.pt
172
+ *.bin
173
+ examples/generated*.gif
174
+ test.ipynb
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: diffusion-model-framework
3
- Version: 1.1.0-rc.1
3
+ Version: 1.1.0-rc.2
4
4
  Project-URL: Documentation, https://www.sachahu.com/docs/diffusion-model-framework/latest/index.html
5
5
  Project-URL: Repository, https://github.com/sachahu1/Diffusion-Model-Framework
6
6
  Author-email: Sacha Hu <sachahu@mindsparkai.com>
@@ -1,4 +1,4 @@
1
1
  [diffend] Oversized file quarantined before diffing.
2
- name: diffusion_model_framework-1.1.0rc1/assets/sampling-small.gif
2
+ name: diffusion_model_framework-1.1.0rc2/assets/sampling-small.gif
3
3
  size: 48830777 bytes
4
4
  sha256: 634d757881475b35905ffb7981d84d813cdc183c09f42c285b25556a4fc6a14f
@@ -0,0 +1 @@
1
+ __version__ = "1.1.0-rc.2"
@@ -0,0 +1,15 @@
1
+ """Hydra-zen config stores for the training example.
2
+
3
+ Importing this package registers every config group (``beta_scheduler``,
4
+ ``diffuser``, ``model``, ``optimizer``, ``dataset``) with hydra-zen's store.
5
+ The top-level configs live next to the scripts that use them, e.g.
6
+ ``examples/config/baseline.yaml``.
7
+ """
8
+
9
+ from diffusion_models._store import ( # noqa: F401
10
+ beta_scheduler,
11
+ dataset,
12
+ diffuser,
13
+ model,
14
+ optimizer,
15
+ )
@@ -0,0 +1,15 @@
1
+ from hydra_zen import builds, store
2
+
3
+ from diffusion_models.gaussian_diffusion.beta_schedulers import (
4
+ CosineBetaScheduler,
5
+ LinearBetaScheduler,
6
+ )
7
+
8
+ beta_scheduler_store = store(group="beta_scheduler")
9
+
10
+ beta_scheduler_store(
11
+ builds(LinearBetaScheduler, populate_full_signature=True), name="linear"
12
+ )
13
+ beta_scheduler_store(
14
+ builds(CosineBetaScheduler, populate_full_signature=True), name="cosine"
15
+ )
@@ -0,0 +1,40 @@
1
+ import torch
2
+ from hydra_zen import builds, store
3
+ from torchvision import datasets
4
+ from torchvision.transforms import v2
5
+
6
+ dataset_store = store(group="dataset")
7
+
8
+ # Normalize(0.5, 0.5) maps [0, 1] -> [-1, 1]; its inverse is mean=-1, std=2.
9
+ ImageTransforms = builds(
10
+ v2.Compose,
11
+ transforms=[
12
+ builds(v2.ToImage),
13
+ builds(v2.Resize, size=["${image_size}", "${image_size}"]),
14
+ builds(v2.ToDtype, dtype=builds(torch.get_default_dtype), scale=True),
15
+ builds(v2.Normalize, mean=[0.5] * 3, std=[0.5] * 3),
16
+ ],
17
+ )
18
+
19
+ dataset_store(
20
+ builds(
21
+ datasets.CelebA,
22
+ root="../data",
23
+ download=False,
24
+ split="train",
25
+ transform=ImageTransforms,
26
+ ),
27
+ name="celeba",
28
+ )
29
+
30
+ # Random images, handy for smoke-testing the pipeline without any data.
31
+ dataset_store(
32
+ builds(
33
+ datasets.FakeData,
34
+ size=64,
35
+ image_size=["${image_channels}", "${image_size}", "${image_size}"],
36
+ transform=ImageTransforms,
37
+ populate_full_signature=True,
38
+ ),
39
+ name="fake",
40
+ )
@@ -0,0 +1,20 @@
1
+ from hydra_zen import builds, store
2
+
3
+ from diffusion_models.gaussian_diffusion.ddimm_diffuser import DdimDiffuser
4
+ from diffusion_models.gaussian_diffusion.gaussian_diffuser import (
5
+ GaussianDiffuser,
6
+ )
7
+
8
+ diffuser_store = store(group="diffuser")
9
+
10
+ diffuser_store(
11
+ builds(
12
+ DdimDiffuser,
13
+ beta_scheduler="${beta_scheduler}",
14
+ populate_full_signature=True,
15
+ ),
16
+ name="ddim",
17
+ )
18
+ diffuser_store(
19
+ builds(GaussianDiffuser, beta_scheduler="${beta_scheduler}"), name="ddpm"
20
+ )
@@ -0,0 +1,14 @@
1
+ from hydra_zen import builds, store
2
+
3
+ from diffusion_models.models.SimpleUnet import SimpleUnet
4
+
5
+ model_store = store(group="model")
6
+
7
+ model_store(
8
+ builds(
9
+ SimpleUnet,
10
+ diffuser="${diffuser}",
11
+ image_channels="${image_channels}",
12
+ ),
13
+ name="simple_unet",
14
+ )
@@ -0,0 +1,30 @@
1
+ from hydra_zen import builds, store
2
+ from torch.optim import AdamW
3
+
4
+ from diffusion_models.utils.adopt import ADOPT
5
+
6
+ optimizer_store = store(group="optimizer")
7
+
8
+ # Partials: the model parameters are only known at runtime.
9
+ # ``hydra_convert="all"`` yields plain tuples/lists instead of OmegaConf
10
+ # containers, keeping optimizer state loadable with ``weights_only=True``.
11
+ optimizer_store(
12
+ builds(
13
+ AdamW,
14
+ lr=2e-4,
15
+ populate_full_signature=True,
16
+ zen_partial=True,
17
+ hydra_convert="all",
18
+ ),
19
+ name="adamw",
20
+ )
21
+ optimizer_store(
22
+ builds(
23
+ ADOPT,
24
+ lr=2e-4,
25
+ populate_full_signature=True,
26
+ zen_partial=True,
27
+ hydra_convert="all",
28
+ ),
29
+ name="adopt",
30
+ )
@@ -1,5 +1,5 @@
1
1
  import pathlib
2
- from typing import Callable, Dict
2
+ from typing import Callable, Dict, Optional, Union
3
3
 
4
4
  import torch
5
5
  from torch.nn import functional as F
@@ -28,6 +28,8 @@ class DiffusionTrainer:
28
28
  log_configuration: LogConfiguration = LogConfiguration(),
29
29
  reverse_transforms: Callable = lambda x: x,
30
30
  device: str = "cuda",
31
+ output_directory: Optional[Union[str, pathlib.Path]] = None,
32
+ run_config: Optional[str] = None,
31
33
  ):
32
34
  """A diffusion trainer framework.
33
35
 
@@ -42,6 +44,12 @@ class DiffusionTrainer:
42
44
  log_configuration: The logging configuration to use.
43
45
  reverse_transforms: The reverse transforms to use.
44
46
  device: The device to use.
47
+ output_directory: Where to write this run's artifacts. When provided,
48
+ checkpoints go to ``<output_directory>/checkpoints`` and tensorboard
49
+ logs to ``<output_directory>/tensorboard``. Otherwise the legacy
50
+ ``../checkpoints/<name>`` and ``../runs/<name>`` locations are used.
51
+ run_config: The resolved configuration (YAML) of the run. It is logged
52
+ to tensorboard and stored in every checkpoint for reproducibility.
45
53
  """
46
54
  self.model = model.to(device)
47
55
  """The diffusion model to use."""
@@ -69,26 +77,40 @@ class DiffusionTrainer:
69
77
  self._image_shape = dataset[0][0].shape
70
78
 
71
79
  self.scaler = torch.amp.GradScaler(
72
- device=device
73
- # init_scale=8192,
80
+ device=device,
81
+ enabled=training_configuration.mixed_precision_training,
74
82
  )
75
83
  """A torch GradScaler object."""
76
84
 
77
85
  self.log_configuration = log_configuration
78
86
  """A LogConfiguration object."""
79
87
 
80
- self.checkpoint_path = (
81
- pathlib.Path("../checkpoints")
82
- / self.training_configuration.training_name
83
- )
88
+ if output_directory is not None:
89
+ output_directory = pathlib.Path(output_directory)
90
+ self.checkpoint_path = output_directory / "checkpoints"
91
+ log_directory = output_directory / "tensorboard"
92
+ else:
93
+ self.checkpoint_path = (
94
+ pathlib.Path("../checkpoints")
95
+ / self.training_configuration.training_name
96
+ )
97
+ log_directory = None
84
98
  """The path to save checkpoints."""
85
99
 
86
- self.checkpoint_path.mkdir(exist_ok=True)
100
+ self.checkpoint_path.mkdir(parents=True, exist_ok=True)
87
101
  self.tensorboard_manager = TensorboardManager(
88
102
  log_name=self.training_configuration.training_name,
103
+ log_directory=log_directory,
89
104
  )
90
105
  """A tensorboard manager instance."""
91
106
 
107
+ self.run_config = run_config
108
+ """The resolved run configuration (YAML), if any."""
109
+ if run_config is not None:
110
+ self.tensorboard_manager.log_text(
111
+ "config", f"```yaml\n{run_config}\n```"
112
+ )
113
+
92
114
  self.reverse_transforms = reverse_transforms
93
115
  """A set of reverse transforms."""
94
116
 
@@ -115,6 +137,7 @@ class DiffusionTrainer:
115
137
  alpha_bars=self.model.diffuser.beta_scheduler.alpha_bars,
116
138
  ),
117
139
  tensorboard_run_name=self.tensorboard_manager.summary_writer.log_dir,
140
+ config_yaml=self.run_config,
118
141
  )
119
142
  checkpoint.to_file(self.checkpoint_path / checkpoint_name)
120
143
 
@@ -179,12 +202,29 @@ class DiffusionTrainer:
179
202
  metrics: A dictionary mapping metric names to values.
180
203
  global_step: The current global step.
181
204
  """
205
+ was_training = self.model.training
182
206
  self.model.eval()
207
+ try:
208
+ self._log_to_tensorboard(metrics=metrics, global_step=global_step)
209
+ finally:
210
+ # Logging switches to eval mode (BatchNorm, etc.); always restore.
211
+ self.model.train(was_training)
212
+
213
+ @torch.no_grad()
214
+ def _log_to_tensorboard(self, metrics: Dict[str, float], global_step: int):
183
215
  if global_step % self.log_configuration.log_rate == 0:
184
216
  self.tensorboard_manager.log_metrics(
185
217
  metrics=metrics, global_step=global_step
186
218
  )
187
219
 
220
+ if (
221
+ self.log_configuration.weight_rate is not None
222
+ and global_step % self.log_configuration.weight_rate == 0
223
+ ):
224
+ self.tensorboard_manager.log_weights(
225
+ model=self.model, global_step=global_step
226
+ )
227
+
188
228
  if (global_step % self.log_configuration.image_rate == 0) and (
189
229
  self.log_configuration.number_of_images > 0
190
230
  ):
@@ -56,7 +56,7 @@ class DdimDiffuser(BaseDiffuser):
56
56
  ** 2
57
57
  ).astype(int)
58
58
  self._time_steps = time_steps + 1
59
- self._time_steps_prev = np.concatenate([[0], time_steps[:-1]])
59
+ self._time_steps_prev = np.concatenate([[0], self._time_steps[:-1]])
60
60
  return list(range(self.number_of_steps))[::-1]
61
61
 
62
62
  def get_timestep(self, number_of_images: int, idx: int) -> Timestep: