loading.mdx 20.1 KB
Newer Older
Patrick von Platen's avatar
Patrick von Platen committed
1
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Nathan Lambert's avatar
Nathan Lambert committed
2
3
4
5
6
7
8
9
10
11
12

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
-->

13
# Load pipelines, models, and schedulers
Patrick von Platen's avatar
Patrick von Platen committed
14

15
16
[[open-in-colab]]

17
Having an easy way to use a diffusion system for inference is essential to 🧨 Diffusers. Diffusion systems often consist of multiple components like parameterized models, tokenizers, and schedulers that interact in complex ways. That is why we designed the [`DiffusionPipeline`] to wrap the complexity of the entire diffusion system into an easy-to-use API, while remaining flexible enough to be adapted for other use cases, such as loading each component individually as building blocks to assemble your own diffusion system.
18

19
Everything you need for inference or training is accessible with the `from_pretrained()` method.
20

21
This guide will show you how to load:
22

23
24
25
26
- pipelines from the Hub and locally
- different components into a pipeline
- checkpoint variants such as different floating point types or non-exponential mean averaged (EMA) weights
- models and schedulers
27

28
29
30
31
32
33
34
35
36
## Diffusion Pipeline

<Tip>

💡 Skip to the [DiffusionPipeline explained](#diffusionpipeline-explained) section if you interested in learning in more detail about how the [`DiffusionPipeline`] class works.

</Tip>

The [`DiffusionPipeline`] class is the simplest and most generic way to load any diffusion model from the [Hub](https://huggingface.co/models?library=diffusers). The [`DiffusionPipeline.from_pretrained`] method automatically detects the correct pipeline class from the checkpoint, downloads and caches all the required configuration and weight files, and returns a pipeline instance ready for inference.
37
38
39
40

```python
from diffusers import DiffusionPipeline

41
42
repo_id = "runwayml/stable-diffusion-v1-5"
pipe = DiffusionPipeline.from_pretrained(repo_id)
43
44
```

45
You can also load a checkpoint with it's specific pipeline class. The example above loaded a Stable Diffusion model; to get the same result, use the [`StableDiffusionPipeline`] class:
46
47

```python
48
49
50
51
52
53
from diffusers import StableDiffusionPipeline

repo_id = "runwayml/stable-diffusion-v1-5"
pipe = StableDiffusionPipeline.from_pretrained(repo_id)
```

54
A checkpoint (such as [`CompVis/stable-diffusion-v1-4`](https://huggingface.co/CompVis/stable-diffusion-v1-4) or [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5)) may also be used for more than one task, like text-to-image or image-to-image. To differentiate what task you want to use the checkpoint for, you have to load it directly with it's corresponding task-specific pipeline class:
55
56
57
58
59
60

```python
from diffusers import StableDiffusionImg2ImgPipeline

repo_id = "runwayml/stable-diffusion-v1-5"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(repo_id)
61
62
```

63
### Local pipeline
64

65
To load a diffusion pipeline locally, use [`git-lfs`](https://git-lfs.github.com/) to manually download the checkpoint (in this case, [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5)) to your local disk. This creates a local folder, `./stable-diffusion-v1-5`, on your disk:
66

67
68
69
70
```bash
git lfs install
git clone https://huggingface.co/runwayml/stable-diffusion-v1-5
```
71

72
Then pass the local path to [`~DiffusionPipeline.from_pretrained`]:
73
74
75
76

```python
from diffusers import DiffusionPipeline

77
repo_id = "./stable-diffusion-v1-5"
78
79
80
stable_diffusion = DiffusionPipeline.from_pretrained(repo_id)
```

81
The [`~DiffusionPipeline.from_pretrained`] method won't download any files from the Hub when it detects a local path, but this also means it won't download and cache the latest changes to a checkpoint.
82

83
### Swap components in a pipeline
84

85
You can customize the default components of any pipeline with another compatible component. Customization is important because:
86

87
88
89
- Changing the scheduler is important for exploring the trade-off between generation speed and quality.
- Different components of a model are typically trained independently and you can swap out a component with a better-performing one.
- During finetuning, usually only some components - like the UNet or text encoder - are trained.
90

91
To find out which schedulers are compatible for customization, you can use the `compatibles` method:
92

93
```py
94
95
96
97
from diffusers import DiffusionPipeline

repo_id = "runwayml/stable-diffusion-v1-5"
stable_diffusion = DiffusionPipeline.from_pretrained(repo_id)
98
stable_diffusion.scheduler.compatibles
99
100
```

101
Let's use the [`SchedulerMixin.from_pretrained`] method to replace the default [`PNDMScheduler`] with a more performant scheduler, [`EulerDiscreteScheduler`]. The `subfolder="scheduler"` argument is required to load the scheduler configuration from the correct [subfolder](https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main/scheduler) of the pipeline repository.
102

103
Then you can pass the new [`EulerDiscreteScheduler`] instance to the `scheduler` argument in [`DiffusionPipeline`]:
104
105
106
107
108
109

```python
from diffusers import DiffusionPipeline, EulerDiscreteScheduler, DPMSolverMultistepScheduler

repo_id = "runwayml/stable-diffusion-v1-5"

110
scheduler = EulerDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
111
112
113
114

stable_diffusion = DiffusionPipeline.from_pretrained(repo_id, scheduler=scheduler)
```

115
### Safety checker
116

117
Diffusion models like Stable Diffusion can generate harmful content, which is why 🧨 Diffusers has a [safety checker](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py) to check generated outputs against known hardcoded NSFW content. If you'd like to disable the safety checker for whatever reason, pass `None` to the `safety_checker` argument:
118
119

```python
120
from diffusers import DiffusionPipeline
121

122
repo_id = "runwayml/stable-diffusion-v1-5"
123
124
125
stable_diffusion = DiffusionPipeline.from_pretrained(repo_id, safety_checker=None)
```

126
127
### Reuse components across pipelines

Steven Liu's avatar
Steven Liu committed
128
You can also reuse the same components in multiple pipelines to avoid loading the weights into RAM twice. Use the [`~DiffusionPipeline.components`] method to save the components:
129
130
131
132
133
134
135
136

```python
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline

model_id = "runwayml/stable-diffusion-v1-5"
stable_diffusion_txt2img = StableDiffusionPipeline.from_pretrained(model_id)

components = stable_diffusion_txt2img.components
137
```
138

139
140
141
Then you can pass the `components` to another pipeline without reloading the weights into RAM:

```py
142
143
144
stable_diffusion_img2img = StableDiffusionImg2ImgPipeline(**components)
```

Steven Liu's avatar
Steven Liu committed
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
You can also pass the components individually to the pipeline if you want more flexibility over which components to reuse or disable. For example, to reuse the same components in the text-to-image pipeline, except for the safety checker and feature extractor, in the image-to-image pipeline:

```py
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline

model_id = "runwayml/stable-diffusion-v1-5"
stable_diffusion_txt2img = StableDiffusionPipeline.from_pretrained(model_id)
stable_diffusion_img2img = StableDiffusionImg2ImgPipeline(
    vae=stable_diffusion_txt2img.vae,
    text_encoder=stable_diffusion_txt2img.text_encoder,
    tokenizer=stable_diffusion_txt2img.tokenizer,
    unet=stable_diffusion_txt2img.unet,
    scheduler=stable_diffusion_txt2img.scheduler,
    safety_checker=None,
    feature_extractor=None,
    requires_safety_checker=False,
)
```

164
## Checkpoint variants
165

166
A checkpoint variant is usually a checkpoint where it's weights are:
167

168
169
- Stored in a different floating point type for lower precision and lower storage, such as [`torch.float16`](https://pytorch.org/docs/stable/tensors.html#data-types), because it only requires half the bandwidth and storage to download. You can't use this variant if you're continuing training or using a CPU.
- Non-exponential mean averaged (EMA) weights which shouldn't be used for inference. You should use these to continue finetuning a model.
170

171
<Tip>
172

173
💡 When the checkpoints have identical model structures, but they were trained on different datasets and with a different training setup, they should be stored in separate repositories instead of variations (for example, [`stable-diffusion-v1-4`] and [`stable-diffusion-v1-5`]).
174

175
</Tip>
176

177
Otherwise, a variant is **identical** to the original checkpoint. They have exactly the same serialization format (like [Safetensors](./using-diffusers/using_safetensors)), model structure, and weights have identical tensor shapes.
178

179
180
181
182
183
| **checkpoint type** | **weight name**                     | **argument for loading weights** |
|---------------------|-------------------------------------|----------------------------------|
| original            | diffusion_pytorch_model.bin         |                                  |
| floating point      | diffusion_pytorch_model.fp16.bin    | `variant`, `torch_dtype`         |
| non-EMA             | diffusion_pytorch_model.non_ema.bin | `variant`                        |
184

185
There are two important arguments to know for loading variants:
186

187
- `torch_dtype` defines the floating point precision of the loaded checkpoints. For example, if you want to save bandwidth by loading a `fp16` variant, you should specify `torch_dtype=torch.float16` to *convert the weights* to `fp16`. Otherwise, the `fp16` weights are converted to the default `fp32` precision. You can also load the original checkpoint without defining the `variant` argument, and convert it to `fp16` with `torch_dtype=torch.float16`. In this case, the default `fp32` weights are downloaded first, and then they're converted to `fp16` after loading.
188

189
- `variant` defines which files should be loaded from the repository. For example, if you want to load a `non_ema` variant from the [`diffusers/stable-diffusion-variants`](https://huggingface.co/diffusers/stable-diffusion-variants/tree/main/unet) repository, you should specify `variant="non_ema"` to download the `non_ema` files.
190

191
192
```python
from diffusers import DiffusionPipeline
193

194
195
196
197
198
199
200
# load fp16 variant
stable_diffusion = DiffusionPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5", variant="fp16", torch_dtype=torch.float16
)
# load non_ema variant
stable_diffusion = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", variant="non_ema")
```
201

202
203
204
To save a checkpoint stored in a different floating point type or as a non-EMA variant, use the [`DiffusionPipeline.save_pretrained`] method and specify the `variant` argument. You should try and save a variant to the same folder as the original checkpoint, so you can load both from the same folder:

```python
205
206
from diffusers import DiffusionPipeline

207
208
209
210
# save as fp16 variant
stable_diffusion.save_pretrained("runwayml/stable-diffusion-v1-5", variant="fp16")
# save as non-ema variant
stable_diffusion.save_pretrained("runwayml/stable-diffusion-v1-5", variant="non_ema")
211
212
```

213
If you don't save the variant to an existing folder, you must specify the `variant` argument otherwise it'll throw an `Exception` because it can't find the original checkpoint:
214

215
216
217
218
219
220
221
```python
# 👎 this won't work
stable_diffusion = DiffusionPipeline.from_pretrained("./stable-diffusion-v1-5", torch_dtype=torch.float16)
# 👍 this works
stable_diffusion = DiffusionPipeline.from_pretrained(
    "./stable-diffusion-v1-5", variant="fp16", torch_dtype=torch.float16
)
222
223
```

224
225
<!--
TODO(Patrick) - Make sure to uncomment this part as soon as things are deprecated.
226

227
#### Using `revision` to load pipeline variants is deprecated
228

229
230
Previously the `revision` argument of [`DiffusionPipeline.from_pretrained`] was heavily used to 
load model variants, e.g.:
231

232
233
```python
from diffusers import DiffusionPipeline
234

235
pipe = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", revision="fp16")
236
237
```

238
However, this behavior is now deprecated since the "revision" argument should (just as it's done in GitHub) better be used to load model checkpoints from a specific commit or branch in development.
239

240
The above example is therefore deprecated and won't be supported anymore for `diffusers >= 1.0.0`.
241

242
<Tip warning={true}>
243

244
245
246
If you load diffusers pipelines or models with `revision="fp16"` or `revision="non_ema"`, 
please make sure to update to code and use `variant="fp16"` or `variation="non_ema"` respectively
instead.
247

248
249
</Tip>
-->
250

251
## Models
252

253
Models are loaded from the [`ModelMixin.from_pretrained`] method, which downloads and caches the latest version of the model weights and configurations. If the latest files are available in the local cache, [`~ModelMixin.from_pretrained`] reuses files in the cache instead of redownloading them.
254

255
Models can be loaded from a subfolder with the `subfolder` argument. For example, the model weights for `runwayml/stable-diffusion-v1-5` are stored in the [`unet`](https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main/unet) subfolder:
256

257
258
259
260
261
```python
from diffusers import UNet2DConditionModel

repo_id = "runwayml/stable-diffusion-v1-5"
model = UNet2DConditionModel.from_pretrained(repo_id, subfolder="unet")
262
263
```

264
Or directly from a repository's [directory](https://huggingface.co/google/ddpm-cifar10-32/tree/main):
265

266
267
268
269
270
```python
from diffusers import UNet2DModel

repo_id = "google/ddpm-cifar10-32"
model = UNet2DModel.from_pretrained(repo_id)
271
272
```

273
You can also load and save model variants by specifying the `variant` argument in [`ModelMixin.from_pretrained`] and [`ModelMixin.save_pretrained`]:
274

275
276
```python
from diffusers import UNet2DConditionModel
277

278
279
280
model = UNet2DConditionModel.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="unet", variant="non-ema")
model.save_pretrained("./local-unet", variant="non-ema")
```
281

282
283
284
## Schedulers

Schedulers are loaded from the [`SchedulerMixin.from_pretrained`] method, and unlike models, schedulers are **not parameterized** or **trained**; they are defined by a configuration file.
285

286
287
Loading schedulers does not consume any significant amount of memory and the same configuration file can be used for a variety of different schedulers.
For example, the following schedulers are compatible with [`StableDiffusionPipeline`] which means you can load the same scheduler configuration file in any of these classes:
288
289

```python
290
291
292
293
294
295
296
297
298
299
from diffusers import StableDiffusionPipeline
from diffusers import (
    DDPMScheduler,
    DDIMScheduler,
    PNDMScheduler,
    LMSDiscreteScheduler,
    EulerDiscreteScheduler,
    EulerAncestralDiscreteScheduler,
    DPMSolverMultistepScheduler,
)
300

301
repo_id = "runwayml/stable-diffusion-v1-5"
302

303
304
305
306
307
308
309
ddpm = DDPMScheduler.from_pretrained(repo_id, subfolder="scheduler")
ddim = DDIMScheduler.from_pretrained(repo_id, subfolder="scheduler")
pndm = PNDMScheduler.from_pretrained(repo_id, subfolder="scheduler")
lms = LMSDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
euler_anc = EulerAncestralDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
euler = EulerDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
dpm = DPMSolverMultistepScheduler.from_pretrained(repo_id, subfolder="scheduler")
310

311
312
313
# replace `dpm` with any of `ddpm`, `ddim`, `pndm`, `lms`, `euler_anc`, `euler`
pipeline = StableDiffusionPipeline.from_pretrained(repo_id, scheduler=dpm)
```
314

315
## DiffusionPipeline explained
316
317
318

As a class method, [`DiffusionPipeline.from_pretrained`] is responsible for two things:

319
320
321
322
- Download the latest version of the folder structure required for inference and cache it. If the latest folder structure is available in the local cache, [`DiffusionPipeline.from_pretrained`] reuses the cache and won't redownload the files.
- Load the cached weights into the correct pipeline [class](./api/pipelines/overview#diffusers-summary) - retrieved from the `model_index.json` file - and return an instance of it.

The pipelines underlying folder structure corresponds directly with their class instances. For example, the [`StableDiffusionPipeline`] corresponds to the folder structure in [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5).
323
324
325
326

```python
from diffusers import DiffusionPipeline

327
repo_id = "runwayml/stable-diffusion-v1-5"
328
329
pipeline = DiffusionPipeline.from_pretrained(repo_id)
print(pipeline)
330
331
```

332
333
334
335
336
337
338
339
340
341
342
You'll see pipeline is an instance of [`StableDiffusionPipeline`], which consists of seven components:

- `"feature_extractor"`: a [`~transformers.CLIPFeatureExtractor`] from 🤗 Transformers.
- `"safety_checker"`: a [component](https://github.com/huggingface/diffusers/blob/e55687e1e15407f60f32242027b7bb8170e58266/src/diffusers/pipelines/stable_diffusion/safety_checker.py#L32) for screening against harmful content.
- `"scheduler"`: an instance of [`PNDMScheduler`].
- `"text_encoder"`: a [`~transformers.CLIPTextModel`] from 🤗 Transformers.
- `"tokenizer"`: a [`~transformers.CLIPTokenizer`] from 🤗 Transformers.
- `"unet"`: an instance of [`UNet2DConditionModel`].
- `"vae"` an instance of [`AutoencoderKL`].

```json
343
344
345
StableDiffusionPipeline {
  "feature_extractor": [
    "transformers",
346
    "CLIPImageProcessor"
347
348
349
350
  ],
  "safety_checker": [
    "stable_diffusion",
    "StableDiffusionSafetyChecker"
351
352
353
  ],
  "scheduler": [
    "diffusers",
354
355
356
357
358
    "PNDMScheduler"
  ],
  "text_encoder": [
    "transformers",
    "CLIPTextModel"
359
360
361
  ],
  "tokenizer": [
    "transformers",
362
    "CLIPTokenizer"
363
364
365
366
367
  ],
  "unet": [
    "diffusers",
    "UNet2DConditionModel"
  ],
368
  "vae": [
369
370
371
372
373
374
    "diffusers",
    "AutoencoderKL"
  ]
}
```

375
Compare the components of the pipeline instance to the [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5) folder structure, and you'll see there is a separate folder for each of the components in the repository:
376
377
378

```
.
379
380
381
382
├── feature_extractor
│   └── preprocessor_config.json
├── model_index.json
├── safety_checker
383
384
385
386
│   ├── config.json
│   └── pytorch_model.bin
├── scheduler
│   └── scheduler_config.json
387
388
389
├── text_encoder
│   ├── config.json
│   └── pytorch_model.bin
390
├── tokenizer
391
│   ├── merges.txt
392
393
│   ├── special_tokens_map.json
│   ├── tokenizer_config.json
394
│   └── vocab.json
395
396
├── unet
│   ├── config.json
397
398
│   ├── diffusion_pytorch_model.bin
└── vae
399
    ├── config.json
400
    ├── diffusion_pytorch_model.bin
401
402
```

403
You can access each of the components of the pipeline as an attribute to view its configuration:
404

405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
```py
pipeline.tokenizer
CLIPTokenizer(
    name_or_path="/root/.cache/huggingface/hub/models--runwayml--stable-diffusion-v1-5/snapshots/39593d5650112b4cc580433f6b0435385882d819/tokenizer",
    vocab_size=49408,
    model_max_length=77,
    is_fast=False,
    padding_side="right",
    truncation_side="right",
    special_tokens={
        "bos_token": AddedToken("<|startoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
        "eos_token": AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
        "unk_token": AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
        "pad_token": "<|endoftext|>",
    },
)
421
```
422
423
424
425
426
427
428
429

Every pipeline expects a `model_index.json` file that tells the [`DiffusionPipeline`]:

- which pipeline class to load from `_class_name`
- which version of 🧨 Diffusers was used to create the model in `_diffusers_version`
- what components from which library are stored in the subfolders (`name` corresponds to the component and subfolder name, `library` corresponds to the name of the library to load the class from, and `class` corresponds to the class name)

```json
430
{
431
432
433
434
  "_class_name": "StableDiffusionPipeline",
  "_diffusers_version": "0.6.0",
  "feature_extractor": [
    "transformers",
435
    "CLIPImageProcessor"
436
437
438
439
  ],
  "safety_checker": [
    "stable_diffusion",
    "StableDiffusionSafetyChecker"
440
441
442
  ],
  "scheduler": [
    "diffusers",
443
444
445
446
447
    "PNDMScheduler"
  ],
  "text_encoder": [
    "transformers",
    "CLIPTextModel"
448
449
450
  ],
  "tokenizer": [
    "transformers",
451
    "CLIPTokenizer"
452
453
454
455
456
  ],
  "unet": [
    "diffusers",
    "UNet2DConditionModel"
  ],
457
  "vae": [
458
459
460
461
    "diffusers",
    "AutoencoderKL"
  ]
}
462
```