File size: 23,244 Bytes
7e77e3e 5704d1f 1ade4d5 5704d1f 300d8bd 5704d1f a02a1d6 5704d1f 03e8165 5704d1f 03e8165 5704d1f 03e8165 5704d1f 03e8165 5704d1f 03e8165 5704d1f 03e8165 5704d1f a02a1d6 5704d1f 03e8165 5704d1f f64c23c 5704d1f 03e8165 5704d1f f64c23c 5704d1f 03e8165 5704d1f f64c23c 5704d1f 03e8165 5704d1f 03e8165 5704d1f 03e8165 5704d1f 66c39b2 5704d1f 956bebb 66c39b2 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 |
---
license: openrail++
language:
- en
library_name: diffusers
tags:
- art
---
# amused
<sup><sub>Images cherry-picked from 512 and 256 models. Images are degraded to load faster. See ./assets/collage_full.png for originals</sub></sup>
📃 Paper: [aMUSEd: An Open MUSE Reproduction](https://huggingface.co/papers/2401.01808)
| Model | Params |
|-------|--------|
| [amused-256](https://huggingface.co/amused/amused-256) | 803M |
| [amused-512](https://huggingface.co/amused/amused-512) | 808M |
Amused is a lightweight text to image model based off of the [muse](https://arxiv.org/pdf/2301.00704.pdf) architecture. Amused is particularly useful in applications that require a lightweight and fast model such as generating many images quickly at once.
*The diagram shows the training and inference pipelines for aMUSEd. aMUSEd consists
of three separately trained components: a pre-trained CLIP-L/14 text encoder, a VQ-GAN, and a
U-ViT. During training, the VQ-GAN encoder maps images to a 16x smaller latent resolution. The
proportion of masked latent tokens is sampled from a cosine masking schedule, e.g. cos(r · π
2 )
with r ∼ Uniform(0, 1). The model is trained via cross-entropy loss to predict the masked tokens.
After the model is trained on 256x256 images, downsampling and upsampling layers are added, and
training is continued on 512x512 images. During inference, the U-ViT is conditioned on the text
encoder’s hidden states and iteratively predicts values for all masked tokens. The cosine masking
schedule determines a percentage of the most confident token predictions to be fixed after every
iteration. After 12 iterations, all tokens have been predicted and are decoded by the VQ-GAN into
image pixels.*
## 1. Usage
### Text to image
#### 256x256 model
```python
import torch
from diffusers import AmusedPipeline
pipe = AmusedPipeline.from_pretrained(
"amused/amused-256", variant="fp16", torch_dtype=torch.float16
)
pipe = pipe.to("cuda")
prompt = "cowboy"
image = pipe(prompt, generator=torch.Generator('cuda').manual_seed(8)).images[0]
image.save('text2image_256.png')
```
#### 512x512 model
```python
import torch
from diffusers import AmusedPipeline
pipe = AmusedPipeline.from_pretrained(
"amused/amused-512", variant="fp16", torch_dtype=torch.float16
)
pipe = pipe.to("cuda")
prompt = "summer in the mountains"
image = pipe(prompt, generator=torch.Generator('cuda').manual_seed(2)).images[0]
image.save('text2image_512.png')
```
### Image to image
#### 256x256 model
```python
import torch
from diffusers import AmusedImg2ImgPipeline
from diffusers.utils import load_image
pipe = AmusedImg2ImgPipeline.from_pretrained(
"amused/amused-256", variant="fp16", torch_dtype=torch.float16
)
pipe = pipe.to("cuda")
prompt = "apple watercolor"
input_image = (
load_image(
"https://raw.githubusercontent.com/huggingface/amused/main/assets/image2image_256_orig.png"
)
.resize((256, 256))
.convert("RGB")
)
image = pipe(prompt, input_image, strength=0.7, generator=torch.Generator('cuda').manual_seed(3)).images[0]
image.save('image2image_256.png')
```
 
#### 512x512 model
```python
import torch
from diffusers import AmusedImg2ImgPipeline
from diffusers.utils import load_image
pipe = AmusedImg2ImgPipeline.from_pretrained(
"amused/amused-512", variant="fp16", torch_dtype=torch.float16
)
pipe = pipe.to("cuda")
prompt = "winter mountains"
input_image = (
load_image(
"https://raw.githubusercontent.com/huggingface/amused/main/assets/image2image_512_orig.png"
)
.resize((512, 512))
.convert("RGB")
)
image = pipe(prompt, input_image, generator=torch.Generator('cuda').manual_seed(15)).images[0]
image.save('image2image_512.png')
```
 
### Inpainting
#### 256x256 model
```python
import torch
from diffusers import AmusedInpaintPipeline
from diffusers.utils import load_image
from PIL import Image
pipe = AmusedInpaintPipeline.from_pretrained(
"amused/amused-256", variant="fp16", torch_dtype=torch.float16
)
pipe = pipe.to("cuda")
prompt = "a man with glasses"
input_image = (
load_image(
"https://raw.githubusercontent.com/huggingface/amused/main/assets/inpainting_256_orig.png"
)
.resize((256, 256))
.convert("RGB")
)
mask = (
load_image(
"https://raw.githubusercontent.com/huggingface/amused/main/assets/inpainting_256_mask.png"
)
.resize((256, 256))
.convert("L")
)
for seed in range(20):
image = pipe(prompt, input_image, mask, generator=torch.Generator('cuda').manual_seed(seed)).images[0]
image.save(f'inpainting_256_{seed}.png')
```
  
#### 512x512 model
```python
import torch
from diffusers import AmusedInpaintPipeline
from diffusers.utils import load_image
pipe = AmusedInpaintPipeline.from_pretrained(
"amused/amused-512", variant="fp16", torch_dtype=torch.float16
)
pipe = pipe.to("cuda")
prompt = "fall mountains"
input_image = (
load_image(
"https://raw.githubusercontent.com/huggingface/amused/main/assets/inpainting_512_orig.jpeg"
)
.resize((512, 512))
.convert("RGB")
)
mask = (
load_image(
"https://raw.githubusercontent.com/huggingface/amused/main/assets/inpainting_512_mask.png"
)
.resize((512, 512))
.convert("L")
)
image = pipe(prompt, input_image, mask, generator=torch.Generator('cuda').manual_seed(0)).images[0]
image.save('inpainting_512.png')
```
## 2. Performance
Amused inherits performance benefits from original [muse](https://arxiv.org/pdf/2301.00704.pdf).
1. Parallel decoding: The model follows a denoising schedule that aims to unmask some percent of tokens at each denoising step. At each step, all masked tokens are predicted, and some number of tokens that the network is most confident about are unmasked. Because multiple tokens are predicted at once, we can generate a full 256x256 or 512x512 image in around 12 steps. In comparison, an autoregressive model must predict a single token at a time. Note that a 256x256 image with the 16x downsampled VAE that muse uses will have 256 tokens.
2. Fewer sampling steps: Compared to many diffusion models, muse requires fewer samples.
Additionally, amused uses the smaller CLIP as its text encoder instead of T5 compared to muse. Amused is also smaller with ~600M params compared the largest 3B param muse model. Note that being smaller, amused produces comparably lower quality results.
### Muse performance knobs
| | Uncompiled Transformer + regular attention | Uncompiled Transformer + flash attention (ms) | Compiled Transformer (ms) | Speed Up |
|---------------------|--------------------------------------------|-------------------------|----------------------|----------|
| 256 Batch Size 1 | 594.7 | 507.7 | 212.1 | 58% |
| 512 Batch Size 1 | 637 | 547 | 249.9 | 54% |
| 256 Batch Size 8 | 719 | 628.6 | 427.8 | 32% |
| 512 Batch Size 8 | 1000 | 917.7 | 703.6 | 23% |
Flash attention is enabled by default in the diffusers codebase through torch `F.scaled_dot_product_attention`
### torch.compile
To use torch.compile, simply wrap the transformer in torch.compile i.e.
```python
pipe.transformer = torch.compile(pipe.transformer)
```
Full snippet:
```python
import torch
from diffusers import AmusedPipeline
pipe = AmusedPipeline.from_pretrained(
"amused/amused-256", variant="fp16", torch_dtype=torch.float16
)
# HERE use torch.compile
pipe.transformer = torch.compile(pipe.transformer)
pipe = pipe.to("cuda")
prompt = "cowboy"
image = pipe(prompt, generator=torch.Generator('cuda').manual_seed(8)).images[0]
image.save('text2image_256.png')
```
## 3. Training
Amused can be finetuned on simple datasets relatively cheaply and quickly. Using 8bit optimizers, lora, and gradient accumulation, amused can be finetuned with as little as 5.5 GB. Here are a set of examples for finetuning amused on some relatively simple datasets. These training recipies are aggressively oriented towards minimal resources and fast verification -- i.e. the batch sizes are quite low and the learning rates are quite high. For optimal quality, you will probably want to increase the batch sizes and decrease learning rates.
All training examples use fp16 mixed precision and gradient checkpointing. We don't show 8 bit adam + lora as its about the same memory use as just using lora (bitsandbytes uses full precision optimizer states for weights below a minimum size).
### Finetuning the 256 checkpoint
These examples finetune on this [nouns](https://huggingface.co/datasets/m1guelpf/nouns) dataset.
Example results:
  
#### Full finetuning
Batch size: 8, Learning rate: 1e-4, Gives decent results in 750-1000 steps
| Batch Size | Gradient Accumulation Steps | Effective Total Batch Size | Memory Used |
|------------|-----------------------------|------------------|-------------|
| 8 | 1 | 8 | 19.7 GB |
| 4 | 2 | 8 | 18.3 GB |
| 1 | 8 | 8 | 17.9 GB |
```sh
accelerate launch training/training.py \
--output_dir <output path> \
--train_batch_size <batch size> \
--gradient_accumulation_steps <gradient accumulation steps> \
--learning_rate 1e-4 \
--pretrained_model_name_or_path amused/amused-256 \
--instance_data_dataset 'm1guelpf/nouns' \
--image_key image \
--prompt_key text \
--resolution 256 \
--mixed_precision fp16 \
--lr_scheduler constant \
--validation_prompts \
'a pixel art character with square red glasses, a baseball-shaped head and a orange-colored body on a dark background' \
'a pixel art character with square orange glasses, a lips-shaped head and a red-colored body on a light background' \
'a pixel art character with square blue glasses, a microwave-shaped head and a purple-colored body on a sunny background' \
'a pixel art character with square red glasses, a baseball-shaped head and a blue-colored body on an orange background' \
'a pixel art character with square red glasses' \
'a pixel art character' \
'square red glasses on a pixel art character' \
'square red glasses on a pixel art character with a baseball-shaped head' \
--max_train_steps 10000 \
--checkpointing_steps 500 \
--validation_steps 250 \
--gradient_checkpointing
```
#### Full finetuning + 8 bit adam
Note that this training config keeps the batch size low and the learning rate high to get results fast with low resources. However, due to 8 bit adam, it will diverge eventually. If you want to train for longer, you will have to up the batch size and lower the learning rate.
Batch size: 16, Learning rate: 2e-5, Gives decent results in ~750 steps
| Batch Size | Gradient Accumulation Steps | Effective Total Batch Size | Memory Used |
|------------|-----------------------------|------------------|-------------|
| 16 | 1 | 16 | 20.1 GB |
| 8 | 2 | 16 | 15.6 GB |
| 1 | 16 | 16 | 10.7 GB |
```sh
accelerate launch training/training.py \
--output_dir <output path> \
--train_batch_size <batch size> \
--gradient_accumulation_steps <gradient accumulation steps> \
--learning_rate 2e-5 \
--use_8bit_adam \
--pretrained_model_name_or_path amused/amused-256 \
--instance_data_dataset 'm1guelpf/nouns' \
--image_key image \
--prompt_key text \
--resolution 256 \
--mixed_precision fp16 \
--lr_scheduler constant \
--validation_prompts \
'a pixel art character with square red glasses, a baseball-shaped head and a orange-colored body on a dark background' \
'a pixel art character with square orange glasses, a lips-shaped head and a red-colored body on a light background' \
'a pixel art character with square blue glasses, a microwave-shaped head and a purple-colored body on a sunny background' \
'a pixel art character with square red glasses, a baseball-shaped head and a blue-colored body on an orange background' \
'a pixel art character with square red glasses' \
'a pixel art character' \
'square red glasses on a pixel art character' \
'square red glasses on a pixel art character with a baseball-shaped head' \
--max_train_steps 10000 \
--checkpointing_steps 500 \
--validation_steps 250 \
--gradient_checkpointing
```
#### Full finetuning + lora
Batch size: 16, Learning rate: 8e-4, Gives decent results in 1000-1250 steps
| Batch Size | Gradient Accumulation Steps | Effective Total Batch Size | Memory Used |
|------------|-----------------------------|------------------|-------------|
| 16 | 1 | 16 | 14.1 GB |
| 8 | 2 | 16 | 10.1 GB |
| 1 | 16 | 16 | 6.5 GB |
```sh
accelerate launch training/training.py \
--output_dir <output path> \
--train_batch_size <batch size> \
--gradient_accumulation_steps <gradient accumulation steps> \
--learning_rate 8e-4 \
--use_lora \
--pretrained_model_name_or_path amused/amused-256 \
--instance_data_dataset 'm1guelpf/nouns' \
--image_key image \
--prompt_key text \
--resolution 256 \
--mixed_precision fp16 \
--lr_scheduler constant \
--validation_prompts \
'a pixel art character with square red glasses, a baseball-shaped head and a orange-colored body on a dark background' \
'a pixel art character with square orange glasses, a lips-shaped head and a red-colored body on a light background' \
'a pixel art character with square blue glasses, a microwave-shaped head and a purple-colored body on a sunny background' \
'a pixel art character with square red glasses, a baseball-shaped head and a blue-colored body on an orange background' \
'a pixel art character with square red glasses' \
'a pixel art character' \
'square red glasses on a pixel art character' \
'square red glasses on a pixel art character with a baseball-shaped head' \
--max_train_steps 10000 \
--checkpointing_steps 500 \
--validation_steps 250 \
--gradient_checkpointing
```
### Finetuning the 512 checkpoint
These examples finetune on this [minecraft](https://huggingface.co/monadical-labs/minecraft-preview) dataset.
Example results:
  
#### Full finetuning
Batch size: 8, Learning rate: 8e-5, Gives decent results in 500-1000 steps
| Batch Size | Gradient Accumulation Steps | Effective Total Batch Size | Memory Used |
|------------|-----------------------------|------------------|-------------|
| 8 | 1 | 8 | 24.2 GB |
| 4 | 2 | 8 | 19.7 GB |
| 1 | 8 | 8 | 16.99 GB |
```sh
accelerate launch training/training.py \
--output_dir <output path> \
--train_batch_size <batch size> \
--gradient_accumulation_steps <gradient accumulation steps> \
--learning_rate 8e-5 \
--pretrained_model_name_or_path amused/amused-512 \
--instance_data_dataset 'monadical-labs/minecraft-preview' \
--prompt_prefix 'minecraft ' \
--image_key image \
--prompt_key text \
--resolution 512 \
--mixed_precision fp16 \
--lr_scheduler constant \
--validation_prompts \
'minecraft Avatar' \
'minecraft character' \
'minecraft' \
'minecraft president' \
'minecraft pig' \
--max_train_steps 10000 \
--checkpointing_steps 500 \
--validation_steps 250 \
--gradient_checkpointing
```
#### Full finetuning + 8 bit adam
Batch size: 8, Learning rate: 5e-6, Gives decent results in 500-1000 steps
| Batch Size | Gradient Accumulation Steps | Effective Total Batch Size | Memory Used |
|------------|-----------------------------|------------------|-------------|
| 8 | 1 | 8 | 21.2 GB |
| 4 | 2 | 8 | 13.3 GB |
| 1 | 8 | 8 | 9.9 GB |
```sh
accelerate launch training/training.py \
--output_dir <output path> \
--train_batch_size <batch size> \
--gradient_accumulation_steps <gradient accumulation steps> \
--learning_rate 5e-6 \
--pretrained_model_name_or_path amused/amused-512 \
--instance_data_dataset 'monadical-labs/minecraft-preview' \
--prompt_prefix 'minecraft ' \
--image_key image \
--prompt_key text \
--resolution 512 \
--mixed_precision fp16 \
--lr_scheduler constant \
--validation_prompts \
'minecraft Avatar' \
'minecraft character' \
'minecraft' \
'minecraft president' \
'minecraft pig' \
--max_train_steps 10000 \
--checkpointing_steps 500 \
--validation_steps 250 \
--gradient_checkpointing
```
#### Full finetuning + lora
Batch size: 8, Learning rate: 1e-4, Gives decent results in 500-1000 steps
| Batch Size | Gradient Accumulation Steps | Effective Total Batch Size | Memory Used |
|------------|-----------------------------|------------------|-------------|
| 8 | 1 | 8 | 12.7 GB |
| 4 | 2 | 8 | 9.0 GB |
| 1 | 8 | 8 | 5.6 GB |
```sh
accelerate launch training/training.py \
--output_dir <output path> \
--train_batch_size <batch size> \
--gradient_accumulation_steps <gradient accumulation steps> \
--learning_rate 1e-4 \
--pretrained_model_name_or_path amused/amused-512 \
--instance_data_dataset 'monadical-labs/minecraft-preview' \
--prompt_prefix 'minecraft ' \
--image_key image \
--prompt_key text \
--resolution 512 \
--mixed_precision fp16 \
--lr_scheduler constant \
--validation_prompts \
'minecraft Avatar' \
'minecraft character' \
'minecraft' \
'minecraft president' \
'minecraft pig' \
--max_train_steps 10000 \
--checkpointing_steps 500 \
--validation_steps 250 \
--gradient_checkpointing
```
### Styledrop
[Styledrop](https://arxiv.org/abs/2306.00983) is an efficient finetuning method for learning a new style from a small number of images. It has an optional first stage to generate human picked additional training samples. The additional training samples can be used to augment the initial images. Our examples exclude the optional additional image selection stage and instead we just finetune on a single image.
This is our example style image:
#### 256
Example results:
  
Learning rate: 4e-4, Gives decent results in 1500-2000 steps
```sh
accelerate launch ./training/training.py \
--output_dir <output path> \
--mixed_precision fp16 \
--report_to wandb \
--use_lora \
--pretrained_model_name_or_path amused/amused-256 \
--train_batch_size 1 \
--lr_scheduler constant \
--learning_rate 4e-4 \
--validation_prompts \
'A chihuahua walking on the street in [V] style' \
'A banana on the table in [V] style' \
'A church on the street in [V] style' \
'A tabby cat walking in the forest in [V] style' \
--instance_data_image './training/A mushroom in [V] style.png' \
--max_train_steps 10000 \
--checkpointing_steps 500 \
--validation_steps 100 \
--resolution 256
```
#### 512
Learning rate: 1e-3, Lora alpha 1, Gives decent results in 1500-2000 steps
Example results:
  
```
accelerate launch ./training/training.py \
--output_dir ../styledrop \
--mixed_precision fp16 \
--report_to wandb \
--use_lora \
--pretrained_model_name_or_path amused/amused-512 \
--train_batch_size 1 \
--lr_scheduler constant \
--learning_rate 1e-3 \
--validation_prompts \
'A chihuahua walking on the street in [V] style' \
'A banana on the table in [V] style' \
'A church on the street in [V] style' \
'A tabby cat walking in the forest in [V] style' \
--instance_data_image './training/A mushroom in [V] style.png' \
--max_train_steps 100000 \
--checkpointing_steps 500 \
--validation_steps 100 \
--resolution 512 \
--lora_alpha 1
```
## 4. Acknowledgements
Suraj led training. William led data and supported training. Patrick supported both training and
data and provided general guidance. Robin trained the VQ-GAN and provided general guidance.
Also, immense thanks to community contributor Isamu Isozaki for helpful discussions and code
contributions.
## 5. Citation
```
@misc{patil2024amused,
title={aMUSEd: An Open MUSE Reproduction},
author={Suraj Patil and William Berman and Robin Rombach and Patrick von Platen},
year={2024},
eprint={2401.01808},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
``` |