Spaces:
Runtime error
Runtime error
File size: 11,920 Bytes
87d40d2 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 |
<!--Copyright 2024 The HuggingFace Team. All rights reserved.
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.
-->
# μ€μΌμ€λ¬
diffusion νμ΄νλΌμΈμ diffusion λͺ¨λΈ, μ€μΌμ€λ¬ λ±μ μ»΄ν¬λνΈλ€λ‘ ꡬμ±λ©λλ€. κ·Έλ¦¬κ³ νμ΄νλΌμΈ μμ μΌλΆ μ»΄ν¬λνΈλ₯Ό λ€λ₯Έ μ»΄ν¬λνΈλ‘ κ΅μ²΄νλ μμ 컀μ€ν°λ§μ΄μ§ μμ κ°λ₯ν©λλ€. μ΄μ κ°μ μ»΄ν¬λνΈ μ»€μ€ν°λ§μ΄μ§μ κ°μ₯ λνμ μΈ μμκ° λ°λ‘ [μ€μΌμ€λ¬](../api/schedulers/overview.md)λ₯Ό κ΅μ²΄νλ κ²μ
λλ€.
μ€μΌμ₯΄λ¬λ λ€μκ³Ό κ°μ΄ diffusion μμ€ν
μ μ λ°μ μΈ λλ
Έμ΄μ§ νλ‘μΈμ€λ₯Ό μ μν©λλ€.
- λλ
Έμ΄μ§ μ€ν
μ μΌλ§λ κ°μ Έκ°μΌ ν κΉ?
- νλ₯ μ μΌλ‘(stochastic) νΉμ νμ μ μΌλ‘(deterministic)?
- λλ
Έμ΄μ§ λ μνμ μ°Ύμλ΄κΈ° μν΄ μ΄λ€ μκ³ λ¦¬μ¦μ μ¬μ©ν΄μΌ ν κΉ?
μ΄λ¬ν νλ‘μΈμ€λ λ€μ λν΄νκ³ , λλ
Έμ΄μ§ μλμ λλ
Έμ΄μ§ νλ¦¬ν° μ¬μ΄μ νΈλ μ΄λ μ€νλ₯Ό μ μν΄μΌ νλ λ¬Έμ κ° λ μ μμ΅λλ€. μ£Όμ΄μ§ νμ΄νλΌμΈμ μ΄λ€ μ€μΌμ€λ¬κ° κ°μ₯ μ ν©νμ§λ₯Ό μ λμ μΌλ‘ νλ¨νλ κ²μ λ§€μ° μ΄λ €μ΄ μΌμ
λλ€. μ΄λ‘ μΈν΄ μΌλ¨ ν΄λΉ μ€μΌμ€λ¬λ₯Ό μ§μ μ¬μ©νμ¬, μμ±λλ μ΄λ―Έμ§λ₯Ό μ§μ λμΌλ‘ 보며, μ μ±μ μΌλ‘ μ±λ₯μ νλ¨ν΄λ³΄λ κ²μ΄ μΆμ²λκ³€ ν©λλ€.
## νμ΄νλΌμΈ λΆλ¬μ€κΈ°
λ¨Όμ μ€ν
μ΄λΈ diffusion νμ΄νλΌμΈμ λΆλ¬μ€λλ‘ ν΄λ³΄κ² μ΅λλ€. λ¬Όλ‘ μ€ν
μ΄λΈ diffusionμ μ¬μ©νκΈ° μν΄μλ, νκΉ
νμ΄μ€ νλΈμ λ±λ‘λ μ¬μ©μμ¬μΌ νλ©°, κ΄λ ¨ [λΌμ΄μΌμ€](https://huggingface.co/runwayml/stable-diffusion-v1-5)μ λμν΄μΌ νλ€λ μ μ μμ§ λ§μμ£ΌμΈμ.
*μμ μ£Ό: λ€λ§, νμ¬ μ κ·λ‘ μμ±ν νκΉ
νμ΄μ€ κ³μ μ λν΄μλ λΌμ΄μΌμ€ λμλ₯Ό μꡬνμ§ μλ κ²μΌλ‘ 보μ
λλ€!*
```python
from huggingface_hub import login
from diffusers import DiffusionPipeline
import torch
# first we need to login with our access token
login()
# Now we can download the pipeline
pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
```
λ€μμΌλ‘, GPUλ‘ μ΄λν©λλ€.
```python
pipeline.to("cuda")
```
## μ€μΌμ€λ¬ μ‘μΈμ€
μ€μΌμ€λ¬λ μΈμ λ νμ΄νλΌμΈμ μ»΄ν¬λνΈλ‘μ μ‘΄μ¬νλ©°, μΌλ°μ μΌλ‘ νμ΄νλΌμΈ μΈμ€ν΄μ€ λ΄μ `scheduler`λΌλ μ΄λ¦μ μμ±(property)μΌλ‘ μ μλμ΄ μμ΅λλ€.
```python
pipeline.scheduler
```
**Output**:
```
PNDMScheduler {
"_class_name": "PNDMScheduler",
"_diffusers_version": "0.8.0.dev0",
"beta_end": 0.012,
"beta_schedule": "scaled_linear",
"beta_start": 0.00085,
"clip_sample": false,
"num_train_timesteps": 1000,
"set_alpha_to_one": false,
"skip_prk_steps": true,
"steps_offset": 1,
"trained_betas": null
}
```
μΆλ ₯ κ²°κ³Όλ₯Ό ν΅ν΄, μ°λ¦¬λ ν΄λΉ μ€μΌμ€λ¬κ° [`PNDMScheduler`]μ μΈμ€ν΄μ€λΌλ κ²μ μ μ μμ΅λλ€. μ΄μ [`PNDMScheduler`]μ λ€λ₯Έ μ€μΌμ€λ¬λ€μ μ±λ₯μ λΉκ΅ν΄λ³΄λλ‘ νκ² μ΅λλ€. λ¨Όμ ν
μ€νΈμ μ¬μ©ν ν둬ννΈλ₯Ό λ€μκ³Ό κ°μ΄ μ μν΄λ³΄λλ‘ νκ² μ΅λλ€.
```python
prompt = "A photograph of an astronaut riding a horse on Mars, high resolution, high definition."
```
λ€μμΌλ‘ μ μ¬ν μ΄λ―Έμ§ μμ±μ 보μ₯νκΈ° μν΄μ, λ€μκ³Ό κ°μ΄ λλ€μλλ₯Ό κ³ μ ν΄μ£Όλλ‘ νκ² μ΅λλ€.
```python
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator).images[0]
image
```
<p align="center">
<br>
<img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_pndm.png" width="400"/>
<br>
</p>
## μ€μΌμ€λ¬ κ΅μ²΄νκΈ°
λ€μμΌλ‘ νμ΄νλΌμΈμ μ€μΌμ€λ¬λ₯Ό λ€λ₯Έ μ€μΌμ€λ¬λ‘ κ΅μ²΄νλ λ°©λ²μ λν΄ μμλ³΄κ² μ΅λλ€. λͺ¨λ μ€μΌμ€λ¬λ [`SchedulerMixin.compatibles`]λΌλ μμ±(property)μ κ°κ³ μμ΅λλ€. ν΄λΉ μμ±μ **νΈν κ°λ₯ν** μ€μΌμ€λ¬λ€μ λν μ 보λ₯Ό λ΄κ³ μμ΅λλ€.
```python
pipeline.scheduler.compatibles
```
**Output**:
```
[diffusers.schedulers.scheduling_lms_discrete.LMSDiscreteScheduler,
diffusers.schedulers.scheduling_ddim.DDIMScheduler,
diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler,
diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler,
diffusers.schedulers.scheduling_pndm.PNDMScheduler,
diffusers.schedulers.scheduling_ddpm.DDPMScheduler,
diffusers.schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteScheduler]
```
νΈνλλ μ€μΌμ€λ¬λ€μ μ΄ν΄λ³΄λ©΄ μλμ κ°μ΅λλ€.
- [`LMSDiscreteScheduler`],
- [`DDIMScheduler`],
- [`DPMSolverMultistepScheduler`],
- [`EulerDiscreteScheduler`],
- [`PNDMScheduler`],
- [`DDPMScheduler`],
- [`EulerAncestralDiscreteScheduler`].
μμ μ μνλ ν둬ννΈλ₯Ό μ¬μ©ν΄μ κ°κ°μ μ€μΌμ€λ¬λ€μ λΉκ΅ν΄λ³΄λλ‘ νκ² μ΅λλ€.
λ¨Όμ νμ΄νλΌμΈ μμ μ€μΌμ€λ¬λ₯Ό λ°κΎΈκΈ° μν΄ [`ConfigMixin.config`] μμ±κ³Ό [`ConfigMixin.from_config`] λ©μλλ₯Ό νμ©ν΄λ³΄λ €κ³ ν©λλ€.
```python
pipeline.scheduler.config
```
**Output**:
```
FrozenDict([('num_train_timesteps', 1000),
('beta_start', 0.00085),
('beta_end', 0.012),
('beta_schedule', 'scaled_linear'),
('trained_betas', None),
('skip_prk_steps', True),
('set_alpha_to_one', False),
('steps_offset', 1),
('_class_name', 'PNDMScheduler'),
('_diffusers_version', '0.8.0.dev0'),
('clip_sample', False)])
```
κΈ°μ‘΄ μ€μΌμ€λ¬μ configλ₯Ό νΈν κ°λ₯ν λ€λ₯Έ μ€μΌμ€λ¬μ μ΄μνλ κ² μμ κ°λ₯ν©λλ€.
λ€μ μμλ κΈ°μ‘΄ μ€μΌμ€λ¬(`pipeline.scheduler`)λ₯Ό λ€λ₯Έ μ’
λ₯μ μ€μΌμ€λ¬(`DDIMScheduler`)λ‘ λ°κΎΈλ μ½λμ
λλ€. κΈ°μ‘΄ μ€μΌμ€λ¬κ° κ°κ³ μλ configλ₯Ό `.from_config` λ©μλμ μΈμλ‘ μ λ¬νλ κ²μ νμΈν μ μμ΅λλ€.
```python
from diffusers import DDIMScheduler
pipeline.scheduler = DDIMScheduler.from_config(pipeline.scheduler.config)
```
μ΄μ νμ΄νλΌμΈμ μ€νν΄μ λ μ€μΌμ€λ¬ μ¬μ΄μ μμ±λ μ΄λ―Έμ§μ ν리ν°λ₯Ό λΉκ΅ν΄λ΄
μλ€.
```python
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator).images[0]
image
```
<p align="center">
<br>
<img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_ddim.png" width="400"/>
<br>
</p>
## μ€μΌμ€λ¬λ€ λΉκ΅ν΄λ³΄κΈ°
μ§κΈκΉμ§λ [`PNDMScheduler`]μ [`DDIMScheduler`] μ€μΌμ€λ¬λ₯Ό μ€νν΄λ³΄μμ΅λλ€. μμ§ λΉκ΅ν΄λ³Ό μ€μΌμ€λ¬λ€μ΄ λ λ§μ΄ λ¨μμμΌλ κ³μ λΉκ΅ν΄λ³΄λλ‘ νκ² μ΅λλ€.
[`LMSDiscreteScheduler`]μ μΌλ°μ μΌλ‘ λ μ’μ κ²°κ³Όλ₯Ό 보μ¬μ€λλ€.
```python
from diffusers import LMSDiscreteScheduler
pipeline.scheduler = LMSDiscreteScheduler.from_config(pipeline.scheduler.config)
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator).images[0]
image
```
<p align="center">
<br>
<img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_lms.png" width="400"/>
<br>
</p>
[`EulerDiscreteScheduler`]μ [`EulerAncestralDiscreteScheduler`] κ³ μ 30λ²μ inference stepλ§μΌλ‘λ λμ ν리ν°μ μ΄λ―Έμ§λ₯Ό μμ±νλ κ²μ μ μ μμ΅λλ€.
```python
from diffusers import EulerDiscreteScheduler
pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator, num_inference_steps=30).images[0]
image
```
<p align="center">
<br>
<img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_euler_discrete.png" width="400"/>
<br>
</p>
```python
from diffusers import EulerAncestralDiscreteScheduler
pipeline.scheduler = EulerAncestralDiscreteScheduler.from_config(pipeline.scheduler.config)
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator, num_inference_steps=30).images[0]
image
```
<p align="center">
<br>
<img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_euler_ancestral.png" width="400"/>
<br>
</p>
μ§κΈ μ΄ λ¬Έμλ₯Ό μμ±νλ νμμ κΈ°μ€μμ , [`DPMSolverMultistepScheduler`]κ° μκ° λλΉ κ°μ₯ μ’μ νμ§μ μ΄λ―Έμ§λ₯Ό μμ±νλ κ² κ°μ΅λλ€. 20λ² μ λμ μ€ν
λ§μΌλ‘λ μ€νλ μ μμ΅λλ€.
```python
from diffusers import DPMSolverMultistepScheduler
pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator, num_inference_steps=20).images[0]
image
```
<p align="center">
<br>
<img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_dpm.png" width="400"/>
<br>
</p>
보μλ€μνΌ μμ±λ μ΄λ―Έμ§λ€μ λ§€μ° λΉμ·νκ³ , λΉμ·ν ν리ν°λ₯Ό 보μ΄λ κ² κ°μ΅λλ€. μ€μ λ‘ μ΄λ€ μ€μΌμ€λ¬λ₯Ό μ νν κ²μΈκ°λ μ’
μ’
νΉμ μ΄μ© μ¬λ‘μ κΈ°λ°ν΄μ κ²°μ λκ³€ ν©λλ€. κ²°κ΅ μ¬λ¬ μ’
λ₯μ μ€μΌμ€λ¬λ₯Ό μ§μ μ€νμμΌλ³΄κ³ λμΌλ‘ μ§μ λΉκ΅ν΄μ νλ¨νλ κ² μ’μ μ νμΌ κ² κ°μ΅λλ€.
## Flaxμμ μ€μΌμ€λ¬ κ΅μ²΄νκΈ°
JAX/Flax μ¬μ©μμΈ κ²½μ° κΈ°λ³Έ νμ΄νλΌμΈ μ€μΌμ€λ¬λ₯Ό λ³κ²½ν μλ μμ΅λλ€. λ€μμ Flax Stable Diffusion νμ΄νλΌμΈκ³Ό μ΄κ³ μ [DDPM-Solver++ μ€μΌμ€λ¬λ₯Ό](../api/schedulers/multistep_dpm_solver) μ¬μ©νμ¬ μΆλ‘ μ μ€ννλ λ°©λ²μ λν μμμ
λλ€ .
```Python
import jax
import numpy as np
from flax.jax_utils import replicate
from flax.training.common_utils import shard
from diffusers import FlaxStableDiffusionPipeline, FlaxDPMSolverMultistepScheduler
model_id = "runwayml/stable-diffusion-v1-5"
scheduler, scheduler_state = FlaxDPMSolverMultistepScheduler.from_pretrained(
model_id,
subfolder="scheduler"
)
pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
model_id,
scheduler=scheduler,
revision="bf16",
dtype=jax.numpy.bfloat16,
)
params["scheduler"] = scheduler_state
# Generate 1 image per parallel device (8 on TPUv2-8 or TPUv3-8)
prompt = "a photo of an astronaut riding a horse on mars"
num_samples = jax.device_count()
prompt_ids = pipeline.prepare_inputs([prompt] * num_samples)
prng_seed = jax.random.PRNGKey(0)
num_inference_steps = 25
# shard inputs and rng
params = replicate(params)
prng_seed = jax.random.split(prng_seed, jax.device_count())
prompt_ids = shard(prompt_ids)
images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
```
<Tip warning={true}>
λ€μ Flax μ€μΌμ€λ¬λ *μμ§* Flax Stable Diffusion νμ΄νλΌμΈκ³Ό νΈνλμ§ μμ΅λλ€.
- `FlaxLMSDiscreteScheduler`
- `FlaxDDPMScheduler`
</Tip>
|