Luis Oala commited on
Commit
d62813f
1 Parent(s): 255a136

first test

Browse files
LICENSE ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ MIT License
2
+
3
+ Copyright (c) 2021 OpenAI
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
README.md CHANGED
@@ -1,5 +1,5 @@
1
  ---
2
- title: Glide
3
  emoji: 🔥
4
  colorFrom: red
5
  colorTo: purple
1
  ---
2
+ title: glide
3
  emoji: 🔥
4
  colorFrom: red
5
  colorTo: purple
README.md~ ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ title: Glide
3
+ emoji: 🔥
4
+ colorFrom: red
5
+ colorTo: purple
6
+ sdk: gradio
7
+ app_file: app.py
8
+ pinned: false
9
+ ---
10
+
11
+ # Configuration
12
+
13
+ `title`: _string_
14
+ Display title for the Space
15
+
16
+ `emoji`: _string_
17
+ Space emoji (emoji-only character allowed)
18
+
19
+ `colorFrom`: _string_
20
+ Color for Thumbnail gradient (red, yellow, green, blue, indigo, purple, pink, gray)
21
+
22
+ `colorTo`: _string_
23
+ Color for Thumbnail gradient (red, yellow, green, blue, indigo, purple, pink, gray)
24
+
25
+ `sdk`: _string_
26
+ Can be either `gradio` or `streamlit`
27
+
28
+ `sdk_version` : _string_
29
+ Only applicable for `streamlit` SDK.
30
+ See [doc](https://hf.co/docs/hub/spaces) for more info on supported versions.
31
+
32
+ `app_file`: _string_
33
+ Path to your main application file (which contains either `gradio` or `streamlit` Python code).
34
+ Path is relative to the root of the repository.
35
+
36
+ `pinned`: _boolean_
37
+ Whether the Space stays on top of your list.
app.py ADDED
@@ -0,0 +1,199 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ os.system('pip install -e .')
3
+ import gradio as gr
4
+
5
+ import base64
6
+ from io import BytesIO
7
+ # from fastapi import FastAPI
8
+
9
+ from PIL import Image
10
+ import torch as th
11
+
12
+ from glide_text2im.download import load_checkpoint
13
+ from glide_text2im.model_creation import (
14
+ create_model_and_diffusion,
15
+ model_and_diffusion_defaults,
16
+ model_and_diffusion_defaults_upsampler
17
+ )
18
+
19
+ # print("Loading models...")
20
+ # app = FastAPI()
21
+
22
+ # This notebook supports both CPU and GPU.
23
+ # On CPU, generating one sample may take on the order of 20 minutes.
24
+ # On a GPU, it should be under a minute.
25
+
26
+ """
27
+ credit: follows the gradio glide example by valhalla https://huggingface.co/spaces/valhalla/glide-text2im
28
+ """
29
+
30
+ has_cuda = th.cuda.is_available()
31
+ device = th.device('cpu' if not has_cuda else 'cuda')
32
+
33
+ # Create base model.
34
+ options = model_and_diffusion_defaults()
35
+ options['use_fp16'] = has_cuda
36
+ options['timestep_respacing'] = '100' # use 100 diffusion steps for fast sampling
37
+ model, diffusion = create_model_and_diffusion(**options)
38
+ model.eval()
39
+ if has_cuda:
40
+ model.convert_to_fp16()
41
+ model.to(device)
42
+ model.load_state_dict(load_checkpoint('base', device))
43
+ print('total base parameters', sum(x.numel() for x in model.parameters()))
44
+
45
+ # Create upsampler model.
46
+ options_up = model_and_diffusion_defaults_upsampler()
47
+ options_up['use_fp16'] = has_cuda
48
+ options_up['timestep_respacing'] = 'fast27' # use 27 diffusion steps for very fast sampling
49
+ model_up, diffusion_up = create_model_and_diffusion(**options_up)
50
+ model_up.eval()
51
+ if has_cuda:
52
+ model_up.convert_to_fp16()
53
+ model_up.to(device)
54
+ model_up.load_state_dict(load_checkpoint('upsample', device))
55
+ print('total upsampler parameters', sum(x.numel() for x in model_up.parameters()))
56
+
57
+
58
+ def get_images(batch: th.Tensor):
59
+ """ Display a batch of images inline. """
60
+ scaled = ((batch + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu()
61
+ reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3])
62
+ return Image.fromarray(reshaped.numpy())
63
+
64
+
65
+ # Create a classifier-free guidance sampling function
66
+ guidance_scale = 3.0
67
+
68
+ def model_fn(x_t, ts, **kwargs):
69
+ half = x_t[: len(x_t) // 2]
70
+ combined = th.cat([half, half], dim=0)
71
+ model_out = model(combined, ts, **kwargs)
72
+ eps, rest = model_out[:, :3], model_out[:, 3:]
73
+ cond_eps, uncond_eps = th.split(eps, len(eps) // 2, dim=0)
74
+ half_eps = uncond_eps + guidance_scale * (cond_eps - uncond_eps)
75
+ eps = th.cat([half_eps, half_eps], dim=0)
76
+ return th.cat([eps, rest], dim=1)
77
+
78
+
79
+ # @app.get("/")
80
+ def read_root():
81
+ return {"glide!"}
82
+
83
+ # @app.get("/{generate}")
84
+ def sample(prompt):
85
+ # Sampling parameters
86
+ batch_size = 1
87
+
88
+ # Tune this parameter to control the sharpness of 256x256 images.
89
+ # A value of 1.0 is sharper, but sometimes results in grainy artifacts.
90
+ upsample_temp = 0.997
91
+
92
+ ##############################
93
+ # Sample from the base model #
94
+ ##############################
95
+
96
+ # Create the text tokens to feed to the model.
97
+ tokens = model.tokenizer.encode(prompt)
98
+ tokens, mask = model.tokenizer.padded_tokens_and_mask(
99
+ tokens, options['text_ctx']
100
+ )
101
+
102
+ # Create the classifier-free guidance tokens (empty)
103
+ full_batch_size = batch_size * 2
104
+ uncond_tokens, uncond_mask = model.tokenizer.padded_tokens_and_mask(
105
+ [], options['text_ctx']
106
+ )
107
+
108
+ # Pack the tokens together into model kwargs.
109
+ model_kwargs = dict(
110
+ tokens=th.tensor(
111
+ [tokens] * batch_size + [uncond_tokens] * batch_size, device=device
112
+ ),
113
+ mask=th.tensor(
114
+ [mask] * batch_size + [uncond_mask] * batch_size,
115
+ dtype=th.bool,
116
+ device=device,
117
+ ),
118
+ )
119
+
120
+ # Sample from the base model.
121
+ model.del_cache()
122
+ samples = diffusion.p_sample_loop(
123
+ model_fn,
124
+ (full_batch_size, 3, options["image_size"], options["image_size"]),
125
+ device=device,
126
+ clip_denoised=True,
127
+ progress=True,
128
+ model_kwargs=model_kwargs,
129
+ cond_fn=None,
130
+ )[:batch_size]
131
+ model.del_cache()
132
+
133
+
134
+ ##############################
135
+ # Upsample the 64x64 samples #
136
+ ##############################
137
+
138
+ tokens = model_up.tokenizer.encode(prompt)
139
+ tokens, mask = model_up.tokenizer.padded_tokens_and_mask(
140
+ tokens, options_up['text_ctx']
141
+ )
142
+
143
+ # Create the model conditioning dict.
144
+ model_kwargs = dict(
145
+ # Low-res image to upsample.
146
+ low_res=((samples+1)*127.5).round()/127.5 - 1,
147
+
148
+ # Text tokens
149
+ tokens=th.tensor(
150
+ [tokens] * batch_size, device=device
151
+ ),
152
+ mask=th.tensor(
153
+ [mask] * batch_size,
154
+ dtype=th.bool,
155
+ device=device,
156
+ ),
157
+ )
158
+
159
+ # Sample from the base model.
160
+ model_up.del_cache()
161
+ up_shape = (batch_size, 3, options_up["image_size"], options_up["image_size"])
162
+ up_samples = diffusion_up.ddim_sample_loop(
163
+ model_up,
164
+ up_shape,
165
+ noise=th.randn(up_shape, device=device) * upsample_temp,
166
+ device=device,
167
+ clip_denoised=True,
168
+ progress=True,
169
+ model_kwargs=model_kwargs,
170
+ cond_fn=None,
171
+ )[:batch_size]
172
+ model_up.del_cache()
173
+
174
+ # Show the output
175
+ image = get_images(up_samples)
176
+ # image = to_base64(image)
177
+ # return {"image": image}
178
+ return image
179
+
180
+
181
+ def to_base64(pil_image):
182
+ buffered = BytesIO()
183
+ pil_image.save(buffered, format="JPEG")
184
+ return base64.b64encode(buffered.getvalue())
185
+
186
+ title = "Interactive demo: glide-text2im"
187
+ description = "Demo for OpenAI's GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models."
188
+ article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2112.10741'>GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models</a> | <a href='https://github.com/openai/glide-text2im/'>Official Repo</a></p>"
189
+ examples =["an oil painting of a corgi"]
190
+
191
+ iface = gr.Interface(fn=sample,
192
+ inputs=gr.inputs.Textbox(label='What would you like to see?'),
193
+ outputs=gr.outputs.Image(type="pil", label="Model input + completions"),
194
+ title=title,
195
+ description=description,
196
+ article=article,
197
+ examples=examples,
198
+ enable_queue=True)
199
+ iface.launch(debug=True)
app.py~ ADDED
@@ -0,0 +1,196 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ import os
3
+ os.system('pip install -e .')
4
+ import gradio as gr
5
+
6
+ import base64
7
+ from io import BytesIO
8
+ # from fastapi import FastAPI
9
+
10
+ from PIL import Image
11
+ import torch as th
12
+
13
+ from glide_text2im.download import load_checkpoint
14
+ from glide_text2im.model_creation import (
15
+ create_model_and_diffusion,
16
+ model_and_diffusion_defaults,
17
+ model_and_diffusion_defaults_upsampler
18
+ )
19
+
20
+ # print("Loading models...")
21
+ # app = FastAPI()
22
+
23
+ # This notebook supports both CPU and GPU.
24
+ # On CPU, generating one sample may take on the order of 20 minutes.
25
+ # On a GPU, it should be under a minute.
26
+
27
+ has_cuda = th.cuda.is_available()
28
+ device = th.device('cpu' if not has_cuda else 'cuda')
29
+
30
+ # Create base model.
31
+ options = model_and_diffusion_defaults()
32
+ options['use_fp16'] = has_cuda
33
+ options['timestep_respacing'] = '100' # use 100 diffusion steps for fast sampling
34
+ model, diffusion = create_model_and_diffusion(**options)
35
+ model.eval()
36
+ if has_cuda:
37
+ model.convert_to_fp16()
38
+ model.to(device)
39
+ model.load_state_dict(load_checkpoint('base', device))
40
+ print('total base parameters', sum(x.numel() for x in model.parameters()))
41
+
42
+ # Create upsampler model.
43
+ options_up = model_and_diffusion_defaults_upsampler()
44
+ options_up['use_fp16'] = has_cuda
45
+ options_up['timestep_respacing'] = 'fast27' # use 27 diffusion steps for very fast sampling
46
+ model_up, diffusion_up = create_model_and_diffusion(**options_up)
47
+ model_up.eval()
48
+ if has_cuda:
49
+ model_up.convert_to_fp16()
50
+ model_up.to(device)
51
+ model_up.load_state_dict(load_checkpoint('upsample', device))
52
+ print('total upsampler parameters', sum(x.numel() for x in model_up.parameters()))
53
+
54
+
55
+ def get_images(batch: th.Tensor):
56
+ """ Display a batch of images inline. """
57
+ scaled = ((batch + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu()
58
+ reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3])
59
+ return Image.fromarray(reshaped.numpy())
60
+
61
+
62
+ # Create a classifier-free guidance sampling function
63
+ guidance_scale = 3.0
64
+
65
+ def model_fn(x_t, ts, **kwargs):
66
+ half = x_t[: len(x_t) // 2]
67
+ combined = th.cat([half, half], dim=0)
68
+ model_out = model(combined, ts, **kwargs)
69
+ eps, rest = model_out[:, :3], model_out[:, 3:]
70
+ cond_eps, uncond_eps = th.split(eps, len(eps) // 2, dim=0)
71
+ half_eps = uncond_eps + guidance_scale * (cond_eps - uncond_eps)
72
+ eps = th.cat([half_eps, half_eps], dim=0)
73
+ return th.cat([eps, rest], dim=1)
74
+
75
+
76
+ # @app.get("/")
77
+ def read_root():
78
+ return {"glide!"}
79
+
80
+ # @app.get("/{generate}")
81
+ def sample(prompt):
82
+ # Sampling parameters
83
+ batch_size = 1
84
+
85
+ # Tune this parameter to control the sharpness of 256x256 images.
86
+ # A value of 1.0 is sharper, but sometimes results in grainy artifacts.
87
+ upsample_temp = 0.997
88
+
89
+ ##############################
90
+ # Sample from the base model #
91
+ ##############################
92
+
93
+ # Create the text tokens to feed to the model.
94
+ tokens = model.tokenizer.encode(prompt)
95
+ tokens, mask = model.tokenizer.padded_tokens_and_mask(
96
+ tokens, options['text_ctx']
97
+ )
98
+
99
+ # Create the classifier-free guidance tokens (empty)
100
+ full_batch_size = batch_size * 2
101
+ uncond_tokens, uncond_mask = model.tokenizer.padded_tokens_and_mask(
102
+ [], options['text_ctx']
103
+ )
104
+
105
+ # Pack the tokens together into model kwargs.
106
+ model_kwargs = dict(
107
+ tokens=th.tensor(
108
+ [tokens] * batch_size + [uncond_tokens] * batch_size, device=device
109
+ ),
110
+ mask=th.tensor(
111
+ [mask] * batch_size + [uncond_mask] * batch_size,
112
+ dtype=th.bool,
113
+ device=device,
114
+ ),
115
+ )
116
+
117
+ # Sample from the base model.
118
+ model.del_cache()
119
+ samples = diffusion.p_sample_loop(
120
+ model_fn,
121
+ (full_batch_size, 3, options["image_size"], options["image_size"]),
122
+ device=device,
123
+ clip_denoised=True,
124
+ progress=True,
125
+ model_kwargs=model_kwargs,
126
+ cond_fn=None,
127
+ )[:batch_size]
128
+ model.del_cache()
129
+
130
+
131
+ ##############################
132
+ # Upsample the 64x64 samples #
133
+ ##############################
134
+
135
+ tokens = model_up.tokenizer.encode(prompt)
136
+ tokens, mask = model_up.tokenizer.padded_tokens_and_mask(
137
+ tokens, options_up['text_ctx']
138
+ )
139
+
140
+ # Create the model conditioning dict.
141
+ model_kwargs = dict(
142
+ # Low-res image to upsample.
143
+ low_res=((samples+1)*127.5).round()/127.5 - 1,
144
+
145
+ # Text tokens
146
+ tokens=th.tensor(
147
+ [tokens] * batch_size, device=device
148
+ ),
149
+ mask=th.tensor(
150
+ [mask] * batch_size,
151
+ dtype=th.bool,
152
+ device=device,
153
+ ),
154
+ )
155
+
156
+ # Sample from the base model.
157
+ model_up.del_cache()
158
+ up_shape = (batch_size, 3, options_up["image_size"], options_up["image_size"])
159
+ up_samples = diffusion_up.ddim_sample_loop(
160
+ model_up,
161
+ up_shape,
162
+ noise=th.randn(up_shape, device=device) * upsample_temp,
163
+ device=device,
164
+ clip_denoised=True,
165
+ progress=True,
166
+ model_kwargs=model_kwargs,
167
+ cond_fn=None,
168
+ )[:batch_size]
169
+ model_up.del_cache()
170
+
171
+ # Show the output
172
+ image = get_images(up_samples)
173
+ # image = to_base64(image)
174
+ # return {"image": image}
175
+ return image
176
+
177
+
178
+ def to_base64(pil_image):
179
+ buffered = BytesIO()
180
+ pil_image.save(buffered, format="JPEG")
181
+ return base64.b64encode(buffered.getvalue())
182
+
183
+ title = "Interactive demo: glide-text2im"
184
+ description = "Demo for OpenAI's GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models."
185
+ article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2112.10741'>GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models</a> | <a href='https://github.com/openai/glide-text2im/'>Official Repo</a></p>"
186
+ examples =["an oil painting of a corgi"]
187
+
188
+ iface = gr.Interface(fn=sample,
189
+ inputs=gr.inputs.Textbox(label='What would you like to see?'),
190
+ outputs=gr.outputs.Image(type="pil", label="Model input + completions"),
191
+ title=title,
192
+ description=description,
193
+ article=article,
194
+ examples=examples,
195
+ enable_queue=True)
196
+ iface.launch(debug=True)
glide_text2im/__init__.py ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ """
2
+ A codebase for performing model inference with a text-conditional diffusion model.
3
+ """
glide_text2im/clip/__init__.py ADDED
File without changes
glide_text2im/clip/attention.py ADDED
@@ -0,0 +1,179 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import math
2
+ from abc import ABC, abstractmethod
3
+ from itertools import product
4
+ from typing import Any, Optional
5
+
6
+ import attr
7
+ import numpy as np
8
+ import torch
9
+
10
+
11
+ @attr.s
12
+ class AttentionMask(ABC):
13
+ query_context_size: int = attr.ib(validator=lambda i, a, x: x >= 1) # type: ignore
14
+ key_context_size: int = attr.ib(validator=lambda i, a, x: x >= 1) # type: ignore
15
+ block_size: int = attr.ib(validator=lambda i, a, x: x >= 1) # type: ignore
16
+ n_head: int = attr.ib(validator=lambda i, a, x: x >= 1) # type: ignore
17
+ is_head_specific: bool = attr.ib(default=False)
18
+ n_query_pad: int = attr.ib(default=0)
19
+ n_key_pad: int = attr.ib(default=0)
20
+
21
+ def __attrs_post_init__(self) -> None:
22
+ if self.query_context_size % self.block_size != 0:
23
+ raise ValueError()
24
+ if self.key_context_size % self.block_size != 0:
25
+ raise ValueError()
26
+ if self.n_query_pad >= self.query_context_size:
27
+ raise ValueError()
28
+ if self.n_key_pad >= self.key_context_size:
29
+ raise ValueError()
30
+
31
+ self.n_query_block = self.query_context_size // self.block_size
32
+ self.n_key_block = self.key_context_size // self.block_size
33
+ self.first_pad_query_block_idx = self.n_query_block - int(
34
+ math.ceil(self.n_query_pad / self.block_size)
35
+ )
36
+ self.first_pad_key_block_idx = self.n_key_block - int(
37
+ math.ceil(self.n_key_pad / self.block_size)
38
+ )
39
+
40
+ def _make_global_layout(self) -> None:
41
+ if not self.is_head_specific:
42
+ m = np.ones([self.n_query_block, self.n_key_block], dtype=np.bool)
43
+ r = product(*[range(n) for n in m.shape])
44
+
45
+ for qb, kb in r:
46
+ m[qb, kb] = np.any(self.block_layout(None, 0, qb, kb, 0))
47
+ else:
48
+ m = np.ones([self.n_head, self.n_query_block, self.n_key_block], dtype=np.bool)
49
+ r = product(*[range(n) for n in m.shape])
50
+
51
+ for h, qb, kb in r:
52
+ m[h, qb, kb] = np.any(self.block_layout(None, h, qb, kb, 0))
53
+
54
+ self.global_layout = m
55
+
56
+ @abstractmethod
57
+ def _block_layout(
58
+ self, blk_shape: Any, head_idx: int, query_idx: int, key_idx: int, blk_idx: int
59
+ ) -> np.ndarray:
60
+ raise NotImplementedError()
61
+
62
+ def block_layout(
63
+ self, blk_shape: Any, head_idx: int, query_idx: int, key_idx: int, blk_idx: int
64
+ ) -> np.ndarray:
65
+ """
66
+ `query_idx`, `key_idx` are block-level, zero-based indices.
67
+ """
68
+
69
+ m = np.ones([self.block_size, self.block_size], dtype=np.bool)
70
+
71
+ if query_idx >= self.first_pad_query_block_idx:
72
+ n_pad = min(
73
+ self.block_size,
74
+ (query_idx + 1) * self.block_size - (self.query_context_size - self.n_query_pad),
75
+ )
76
+ assert n_pad > 0
77
+ m[self.block_size - n_pad :] = False
78
+ if key_idx >= self.first_pad_key_block_idx:
79
+ n_pad = min(
80
+ self.block_size,
81
+ (key_idx + 1) * self.block_size - (self.key_context_size - self.n_key_pad),
82
+ )
83
+ assert n_pad > 0
84
+ m[:, self.block_size - n_pad :] = False
85
+
86
+ return m & self._block_layout(blk_shape, head_idx, query_idx, key_idx, blk_idx)
87
+
88
+
89
+ @attr.s
90
+ class DenseAttentionMask(AttentionMask):
91
+ def __attrs_post_init__(self) -> None:
92
+ super().__attrs_post_init__()
93
+
94
+ self.global_layout = np.ones([self.n_query_block, self.n_key_block], dtype=np.bool)
95
+ n_zero_query_blocks = self.n_query_pad // self.block_size
96
+ n_zero_key_blocks = self.n_key_pad // self.block_size
97
+ self.global_layout[self.n_query_block - n_zero_query_blocks :] = False
98
+ self.global_layout[:, self.n_key_block - n_zero_key_blocks :] = False
99
+
100
+ def _block_layout(
101
+ self, blk_shape: Any, head_idx: int, query_idx: int, key_idx: int, blk_idx: int
102
+ ) -> np.ndarray:
103
+ return np.ones([self.block_size, self.block_size], dtype=np.bool)
104
+
105
+
106
+ @attr.s
107
+ class DenseCausalAttentionMask(AttentionMask):
108
+ def __attrs_post_init__(self) -> None:
109
+ super().__attrs_post_init__()
110
+
111
+ self.global_layout = np.tril(np.ones([self.n_query_block, self.n_key_block], dtype=np.bool))
112
+ n_zero_query_blocks = self.n_query_pad // self.block_size
113
+ n_zero_key_blocks = self.n_key_pad // self.block_size
114
+ self.global_layout[self.n_query_block - n_zero_query_blocks :] = False
115
+ self.global_layout[:, self.n_key_block - n_zero_key_blocks :] = False
116
+
117
+ def _block_layout(
118
+ self, blk_shape: Any, head_idx: int, query_idx: int, key_idx: int, blk_idx: int
119
+ ) -> np.ndarray:
120
+ if query_idx > key_idx:
121
+ return np.ones(2 * [self.block_size], dtype=np.bool)
122
+ elif query_idx < key_idx:
123
+ return np.zeros(2 * [self.block_size], dtype=np.bool)
124
+ else:
125
+ return np.tril(np.ones(2 * [self.block_size], dtype=np.bool))
126
+
127
+
128
+ @attr.s(eq=False, repr=False)
129
+ class AttentionInfo:
130
+ n_heads: int = attr.ib()
131
+ ctx_blks_q: int = attr.ib()
132
+ ctx_blks_k: int = attr.ib()
133
+ block_size: int = attr.ib()
134
+ pytorch_attn_bias: Optional[torch.Tensor] = attr.ib()
135
+
136
+
137
+ def to_attention_info(d: AttentionMask) -> AttentionInfo:
138
+ return AttentionInfo(
139
+ n_heads=d.n_head,
140
+ ctx_blks_q=d.n_query_block,
141
+ ctx_blks_k=d.n_key_block,
142
+ block_size=d.block_size,
143
+ pytorch_attn_bias=None,
144
+ )
145
+
146
+
147
+ def make_full_layout(d: AttentionMask) -> np.ndarray:
148
+ """
149
+ Returns the `context_size x context_size` layout matrix described by `d`. If the layout is dependent on the index of
150
+ the attention head, a `attention_head x context_size x context_size` layout matrix is returned instead.
151
+ """
152
+
153
+ if not d.is_head_specific:
154
+ u = np.reshape(d.global_layout, [d.n_query_block, d.n_key_block, 1, 1])
155
+ r = product(range(d.n_query_block), range(d.n_key_block))
156
+ v = np.array([d.block_layout(None, 0, i, j, 0) for i, j in r])
157
+ v = np.reshape(v, [d.n_query_block, d.n_key_block, d.block_size, d.block_size])
158
+
159
+ w = u * v
160
+ w = np.transpose(w, [0, 2, 1, 3])
161
+ w = np.reshape(w, [d.query_context_size, d.key_context_size])
162
+ return w
163
+ else:
164
+ if len(d.global_layout.shape) == 2:
165
+ u = np.reshape(d.global_layout, [1, d.n_query_block, d.n_key_block, 1, 1])
166
+ u = np.tile(u, [d.n_head, 1, 1, 1, 1])
167
+ elif len(d.global_layout.shape) == 3:
168
+ u = np.reshape(d.global_layout, [d.n_head, d.n_query_block, d.n_key_block, 1, 1])
169
+ else:
170
+ raise RuntimeError()
171
+
172
+ s = product(range(d.n_head), range(d.n_query_block), range(d.n_key_block))
173
+ v = np.array([d.block_layout(None, i, j, k, 0) for i, j, k in s])
174
+ v = np.reshape(v, [d.n_head, d.n_query_block, d.n_key_block, d.block_size, d.block_size])
175
+
176
+ w = u * v
177
+ w = np.transpose(w, [0, 1, 3, 2, 4])
178
+ w = np.reshape(w, [d.n_head, d.query_context_size, d.key_context_size])
179
+ return w
glide_text2im/clip/config.yaml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ logit_scale: 100.0
2
+
3
+ # Diffusion settings
4
+ beta_schedule: "squaredcos_cap_v2"
5
+ n_timesteps: 1000
6
+
7
+ # Architecture settings
8
+ image_size: 64
9
+ patch_size: 4
10
+ n_vocab: 65536
11
+ max_text_len: 77
12
+ n_embd: 512
13
+ n_head_state_text: 64
14
+ n_head_text: 8
15
+ n_xf_blocks_text: 12
16
+ n_head_state_image: 64
17
+ n_head_image: 12
18
+ n_xf_blocks_image: 12
glide_text2im/clip/encoders.py ADDED
@@ -0,0 +1,497 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import math
2
+ from collections import OrderedDict
3
+ from typing import List, Optional, Tuple, cast
4
+
5
+ import attr
6
+ import numpy as np
7
+ import torch
8
+ import torch.nn as nn
9
+ import torch.nn.functional as F
10
+
11
+ from .attention import (
12
+ AttentionInfo,
13
+ DenseAttentionMask,
14
+ DenseCausalAttentionMask,
15
+ make_full_layout,
16
+ to_attention_info,
17
+ )
18
+ from .utils import Affine, LayerNorm, zero_key_bias_grad
19
+
20
+ # Constants used in the original CLIP implementation.
21
+ image_channel_means = [122.77093945, 116.74601272, 104.09373519]
22
+ image_channel_stds = [68.50053285, 66.63215831, 70.32316309]
23
+
24
+
25
+ @attr.s(eq=False, repr=False)
26
+ class TextEmbedding(nn.Module):
27
+ n_vocab: int = attr.ib()
28
+ n_context: int = attr.ib()
29
+ n_state: int = attr.ib()
30
+ device: torch.device = attr.ib(default=torch.device("cuda"))
31
+
32
+ def __attrs_post_init__(self) -> None:
33
+ super().__init__()
34
+
35
+ w_voc = torch.empty((self.n_vocab, self.n_state), dtype=torch.float32, device=self.device)
36
+ w_pos = torch.empty((self.n_context, self.n_state), dtype=torch.float32, device=self.device)
37
+
38
+ with torch.no_grad():
39
+ w_voc.normal_(std=0.02)
40
+ w_pos.normal_(std=0.01)
41
+
42
+ self.w_voc = nn.Parameter(w_voc)
43
+ self.w_pos = nn.Parameter(w_pos)
44
+
45
+ def forward(self, x: torch.Tensor) -> torch.Tensor:
46
+ if len(x.shape) != 2:
47
+ raise ValueError()
48
+
49
+ return F.embedding(x, self.w_voc) + self.w_pos[None, :, :]
50
+
51
+
52
+ @attr.s(eq=False, repr=False)
53
+ class ImageEmbedding(nn.Module):
54
+ image_size: int = attr.ib()
55
+ patch_size: int = attr.ib()
56
+ n_state: int = attr.ib()
57
+ n_timestep: int = attr.ib(default=0)
58
+ device: torch.device = attr.ib(default=torch.device("cuda"))
59
+
60
+ def __attrs_post_init__(self) -> None:
61
+ super().__init__()
62
+
63
+ if self.image_size % self.patch_size != 0:
64
+ raise ValueError()
65
+
66
+ n_patch = self.image_size // self.patch_size
67
+ patch_proj = torch.empty(
68
+ (self.n_state, 3) + 2 * (self.patch_size,), dtype=torch.float32, device=self.device
69
+ )
70
+ w_pos = torch.empty(
71
+ (1 + n_patch ** 2, self.n_state), dtype=torch.float32, device=self.device
72
+ )
73
+
74
+ with torch.no_grad():
75
+ if self.n_timestep == 0:
76
+ pred_state = torch.empty((self.n_state,), dtype=torch.float32, device=self.device)
77
+ pred_state.normal_(std=1 / np.sqrt(self.n_state))
78
+ self.pred_state = nn.Parameter(pred_state)
79
+ else:
80
+ w_t = torch.empty(
81
+ (self.n_timestep, self.n_state), dtype=torch.float32, device=self.device
82
+ )
83
+ w_t.normal_(std=1 / np.sqrt(self.n_state))
84
+ self.w_t = nn.Parameter(w_t)
85
+
86
+ patch_proj.normal_(std=np.sqrt(2 / (self.n_state * self.patch_size ** 2)))
87
+ w_pos.normal_(std=1 / np.sqrt(self.n_state))
88
+
89
+ self.patch_proj = nn.Parameter(patch_proj)
90
+ self.w_pos = nn.Parameter(w_pos)
91
+
92
+ self.channel_means = torch.tensor(
93
+ image_channel_means, dtype=torch.float32, device=self.device
94
+ )[None, :, None, None]
95
+ self.channel_stds = torch.tensor(
96
+ image_channel_stds, dtype=torch.float32, device=self.device
97
+ )[None, :, None, None]
98
+ self.ln = LayerNorm(self.n_state, eps=1e-5, device=self.device)
99
+
100
+ def forward(self, x: torch.Tensor, t: Optional[torch.Tensor] = None) -> torch.Tensor:
101
+ if len(x.shape) != 4:
102
+ raise ValueError("input should be 4d")
103
+ if x.shape[1] != 3:
104
+ raise ValueError("input should have 3 channels")
105
+ if not (x.shape[2] == self.image_size and x.shape[3] == self.image_size):
106
+ raise ValueError(f"input is not {self.image_size} x {self.image_size}")
107
+
108
+ if (self.n_timestep == 0 and t is not None) or (self.n_timestep != 0 and t is None):
109
+ raise ValueError()
110
+ if self.n_timestep != 0:
111
+ assert t is not None
112
+ if len(t.shape) != 1:
113
+ raise ValueError()
114
+ if t.shape[0] != x.shape[0]:
115
+ raise ValueError()
116
+
117
+ x = (x - self.channel_means) / self.channel_stds
118
+ x = F.conv2d(x, self.patch_proj, stride=self.patch_size)
119
+ x = x.reshape(x.shape[0], self.n_state, (self.image_size // self.patch_size) ** 2).permute(
120
+ 0, 2, 1
121
+ )
122
+
123
+ sot = (
124
+ self.pred_state[None, None].expand(x.shape[0], -1, -1)
125
+ if self.n_timestep == 0
126
+ else F.embedding(cast(torch.Tensor, t), self.w_t)[:, None]
127
+ )
128
+ x = torch.cat((sot, x), dim=1) + self.w_pos[None]
129
+ return self.ln(x)
130
+
131
+
132
+ @attr.s(eq=False, repr=False)
133
+ class AttentionResblock(nn.Module):
134
+ n_state: int = attr.ib()
135
+ n_resblocks: int = attr.ib()
136
+ attn_fn: AttentionInfo = attr.ib()
137
+ device: torch.device = attr.ib(default=torch.device("cuda"))
138
+
139
+ def __attrs_post_init__(self) -> None:
140
+ super().__init__()
141
+
142
+ self.n_head_state = self.n_state // self.attn_fn.n_heads
143
+ self.qk_scale = 1 / np.sqrt(self.n_head_state)
144
+
145
+ self.ln = LayerNorm(self.n_state, eps=1e-5, device=self.device)
146
+ self.f_q = Affine(
147
+ self.n_state,
148
+ self.n_state,
149
+ std=1 / math.sqrt(self.n_state),
150
+ use_bias=True,
151
+ bias_filter_fn=zero_key_bias_grad,
152
+ device=self.device,
153
+ )
154
+ self.f_k = Affine(
155
+ self.n_state,
156
+ self.n_state,
157
+ std=1 / math.sqrt(self.n_state),
158
+ use_bias=False,
159
+ bias_filter_fn=zero_key_bias_grad,
160
+ device=self.device,
161
+ )
162
+ self.f_v = Affine(
163
+ self.n_state,
164
+ self.n_state,
165
+ std=1 / math.sqrt(self.n_state),
166
+ use_bias=True,
167
+ bias_filter_fn=zero_key_bias_grad,
168
+ device=self.device,
169
+ )
170
+ self.f_c = Affine(
171
+ self.n_state,
172
+ self.n_state,
173
+ use_bias=True,
174
+ std=1 / np.sqrt(self.n_state * self.n_resblocks ** 2),
175
+ device=self.device,
176
+ ) # XXX
177
+
178
+ def forward(self, m: torch.Tensor) -> torch.Tensor:
179
+ n_context = m.shape[1]
180
+ n_query_pad = self.attn_fn.ctx_blks_q * self.attn_fn.block_size - n_context
181
+ n_key_pad = self.attn_fn.ctx_blks_k * self.attn_fn.block_size - n_context
182
+ assert n_query_pad >= 0
183
+ assert n_key_pad >= 0
184
+
185
+ r = m
186
+ r = self.ln(r)
187
+ q, k, v = self.f_q(r), self.f_k(r), self.f_v(r)
188
+
189
+ if n_query_pad != 0:
190
+ q = F.pad(q, (0, 0, 0, n_query_pad))
191
+
192
+ if n_key_pad != 0:
193
+ k = F.pad(k, (0, 0, 0, n_key_pad))
194
+ v = F.pad(v, (0, 0, 0, n_key_pad))
195
+
196
+ q = q.view([q.shape[0], -1, self.attn_fn.n_heads, self.n_head_state]).permute((0, 2, 1, 3))
197
+ k = k.view([k.shape[0], -1, self.attn_fn.n_heads, self.n_head_state]).permute((0, 2, 1, 3))
198
+ v = v.view([v.shape[0], -1, self.attn_fn.n_heads, self.n_head_state]).permute((0, 2, 1, 3))
199
+ w = torch.einsum(
200
+ "bhcd,bhkd->bhck", q * math.sqrt(self.qk_scale), k * math.sqrt(self.qk_scale)
201
+ )
202
+
203
+ if hasattr(self.attn_fn, "pytorch_attn_bias"):
204
+ bias = self.attn_fn.pytorch_attn_bias
205
+ assert len(bias.shape) in {2, 3}
206
+
207
+ if len(bias.shape) == 2:
208
+ w = torch.softmax(w + self.attn_fn.pytorch_attn_bias[None, None], dim=-1)
209
+ elif len(bias.shape) == 3:
210
+ w = torch.softmax(w + self.attn_fn.pytorch_attn_bias[None], dim=-1)
211
+ else:
212
+ w = torch.softmax(w, dim=-1)
213
+
214
+ r = torch.einsum("bhck,bhkd->bhcd", w, v)
215
+ r = r.permute((0, 2, 1, 3)).reshape((r.shape[0], -1, self.n_state))
216
+
217
+ if n_query_pad != 0:
218
+ r = r[:, :-n_query_pad]
219
+
220
+ assert r.shape[1] == n_context
221
+
222
+ r = self.f_c(r)
223
+ return m + r
224
+
225
+
226
+ @attr.s(eq=False, repr=False)
227
+ class FullyConnectedResblock(nn.Module):
228
+ """
229
+ Not imported from other files because we retain Alec's original inits.
230
+ """
231
+
232
+ n_state: int = attr.ib()
233
+ n_resblocks: int = attr.ib()
234
+ device: torch.device = attr.ib(default=torch.device("cuda"))
235
+
236
+ def __attrs_post_init__(self) -> None:
237
+ super().__init__()
238
+
239
+ self.ln = LayerNorm(self.n_state, eps=1e-5, device=self.device)
240
+ self.f_1 = Affine(
241
+ self.n_state,
242
+ 4 * self.n_state,
243
+ use_bias=True,
244
+ std=np.sqrt(2 / (4 * self.n_state)),
245
+ device=self.device,
246
+ )
247
+ self.f_2 = Affine(
248
+ 4 * self.n_state,
249
+ self.n_state,
250
+ use_bias=True,
251
+ std=1 / np.sqrt(self.n_state * self.n_resblocks ** 2),
252
+ device=self.device,
253
+ ) # XXX
254
+
255
+ def forward(self, m: torch.Tensor) -> torch.Tensor:
256
+ r = m
257
+ r = self.ln(r)
258
+
259
+ r = self.f_2(F.gelu(self.f_1(r)))
260
+ return m + r
261
+
262
+
263
+ @attr.s(eq=False, repr=False)
264
+ class TransformerBlock(nn.Module):
265
+ n_state: int = attr.ib()
266
+ n_resblocks: int = attr.ib()
267
+ attn_fn: AttentionInfo = attr.ib()
268
+ device: torch.device = attr.ib(default=torch.device("cuda"))
269
+
270
+ def __attrs_post_init__(self) -> None:
271
+ super().__init__()
272
+
273
+ self.f_attn = AttentionResblock(
274
+ self.n_state,
275
+ self.n_resblocks,
276
+ self.attn_fn,
277
+ self.device,
278
+ )
279
+ self.f_mlp = FullyConnectedResblock(self.n_state, self.n_resblocks, self.device)
280
+
281
+ def forward(self, x: torch.Tensor) -> torch.Tensor:
282
+ return self.f_mlp(self.f_attn(x))
283
+
284
+
285
+ @attr.s(eq=False, repr=False)
286
+ class TextFeatureExtractor(nn.Module):
287
+ n_state: int = attr.ib()
288
+ n_embd: int = attr.ib()
289
+ device: torch.device = attr.ib(default=torch.device("cuda"))
290
+
291
+ def __attrs_post_init__(self) -> None:
292
+ super().__init__()
293
+
294
+ self.ln = LayerNorm(self.n_state, eps=1e-5, device=self.device)
295
+ self.f = Affine(self.n_state, self.n_embd, use_bias=False, device=self.device)
296
+
297
+ def forward(
298
+ self, text: torch.Tensor, text_len: torch.Tensor, return_probe_features: bool = False
299
+ ) -> torch.Tensor:
300
+ if len(text.shape) != 3:
301
+ raise ValueError("expected text to be 3d")
302
+ if len(text_len.shape) != 1:
303
+ raise ValueError("expected text length to be 1d")
304
+ if text.shape[0] != text_len.shape[0]:
305
+ raise ValueError("text and text_len have inconsistent batch dimensions")
306
+
307
+ index = (text_len - 1)[:, None, None].expand(-1, 1, text.shape[2])
308
+ x = torch.gather(text, dim=1, index=index)
309
+ assert list(x.shape) == [text.shape[0], 1, text.shape[2]]
310
+
311
+ if return_probe_features:
312
+ return x[:, 0]
313
+
314
+ x = self.ln(x)
315
+ return self.f(x[:, 0])
316
+
317
+
318
+ @attr.s(eq=False, repr=False)
319
+ class ImageFeatureExtractor(nn.Module):
320
+ n_state: int = attr.ib()
321
+ n_embd: int = attr.ib()
322
+ device: torch.device = attr.ib(default=torch.device("cuda"))
323
+
324
+ def __attrs_post_init__(self) -> None:
325
+ super().__init__()
326
+
327
+ self.ln = LayerNorm(self.n_state, eps=1e-5, device=self.device)
328
+ self.f = Affine(self.n_state, self.n_embd, use_bias=False, device=self.device)
329
+
330
+ def forward(self, x: torch.Tensor, return_probe_features: bool = False) -> torch.Tensor:
331
+ if return_probe_features:
332
+ return x[:, 0]
333
+
334
+ x = self.ln(x[:, :1])
335
+ return self.f(x[:, 0])
336
+
337
+
338
+ @attr.s(eq=False, repr=False)
339
+ class TextEncoder(nn.Module):
340
+ n_bpe_vocab: int = attr.ib()
341
+ max_text_len: int = attr.ib()
342
+ n_embd: int = attr.ib()
343
+ n_head: int = attr.ib()
344
+ n_xf_blocks: int = attr.ib()
345
+ n_head_state: int = attr.ib(default=64)
346
+ device: torch.device = attr.ib(default=torch.device("cuda"))
347
+ block_size: int = attr.ib(init=False, default=32)
348
+
349
+ def __attrs_post_init__(self) -> None:
350
+ super().__init__()
351
+
352
+ self.n_state = self.n_head * self.n_head_state
353
+ n_rounded_context = self.block_size * int(math.ceil(self.max_text_len / self.block_size))
354
+ n_pad = n_rounded_context - self.max_text_len
355
+
356
+ args = (
357
+ n_rounded_context,
358
+ n_rounded_context,
359
+ self.block_size,
360
+ self.n_head,
361
+ False,
362
+ n_pad,
363
+ n_pad,
364
+ )
365
+ mask = DenseCausalAttentionMask(*args)
366
+ attn_fn = to_attention_info(mask)
367
+
368
+ m = 1 - make_full_layout(mask).astype(np.float32)
369
+ m[m == 1] = -1e10
370
+ attn_fn.pytorch_attn_bias = torch.from_numpy(m).to(self.device)
371
+
372
+ blocks: List[Tuple[str, nn.Module]] = [
373
+ (
374
+ "input",
375
+ TextEmbedding(
376
+ self.n_bpe_vocab, self.max_text_len, self.n_state, device=self.device
377
+ ),
378
+ )
379
+ ]
380
+
381
+ for i in range(self.n_xf_blocks):
382
+ blocks.append(
383
+ (
384
+ f"block_{i}",
385
+ TransformerBlock(self.n_state, 2 * self.n_xf_blocks, attn_fn, self.device),
386
+ )
387
+ )
388
+
389
+ blocks.append(
390
+ ("output", TextFeatureExtractor(self.n_state, self.n_embd, device=self.device))
391
+ )
392
+
393
+ self.blocks = nn.ModuleDict(OrderedDict(blocks))
394
+
395
+ def forward(
396
+ self,
397
+ text: torch.Tensor,
398
+ text_len: torch.Tensor,
399
+ return_probe_features: bool = False,
400
+ ) -> torch.Tensor:
401
+
402
+ n_batch = text.shape[0]
403
+ h = self.blocks["input"](text)
404
+
405
+ for i in range(self.n_xf_blocks):
406
+ h = self.blocks[f"block_{i}"](h)
407
+
408
+ h = self.blocks["output"](h, text_len, return_probe_features=return_probe_features)
409
+
410
+ assert list(h.shape) == [
411
+ n_batch,
412
+ self.n_embd if not return_probe_features else self.n_state,
413
+ ]
414
+ return h
415
+
416
+
417
+ @attr.s(eq=False, repr=False)
418
+ class ImageEncoder(nn.Module):
419
+ image_size: int = attr.ib()
420
+ patch_size: int = attr.ib()
421
+ n_embd: int = attr.ib()
422
+ n_head: int = attr.ib()
423
+ n_xf_blocks: int = attr.ib()
424
+ n_head_state: int = attr.ib(default=64)
425
+ n_timestep: int = attr.ib(default=0)
426
+ device: torch.device = attr.ib(default=torch.device("cuda"))
427
+ block_size: int = attr.ib(init=False, default=32)
428
+
429
+ def __attrs_post_init__(self) -> None:
430
+ super().__init__()
431
+
432
+ self.n_state = self.n_head * self.n_head_state
433
+ self.n_context = 1 + (self.image_size // self.patch_size) ** 2
434
+ n_rounded_context = self.block_size * int(math.ceil(self.n_context / self.block_size))
435
+ n_pad = n_rounded_context - self.n_context
436
+
437
+ args = (
438
+ n_rounded_context,
439
+ n_rounded_context,
440
+ self.block_size,
441
+ self.n_head,
442
+ False,
443
+ n_pad,
444
+ n_pad,
445
+ )
446
+ mask = DenseAttentionMask(*args)
447
+ attn_fn = to_attention_info(mask)
448
+
449
+ m = 1 - make_full_layout(mask).astype(np.float32)
450
+ m[m == 1] = -1e10
451
+ attn_fn.pytorch_attn_bias = torch.from_numpy(m).to(self.device)
452
+
453
+ blocks: List[Tuple[str, nn.Module]] = [
454
+ (
455
+ "input",
456
+ ImageEmbedding(
457
+ self.image_size,
458
+ self.patch_size,
459
+ self.n_state,
460
+ n_timestep=self.n_timestep,
461
+ device=self.device,
462
+ ),
463
+ )
464
+ ]
465
+
466
+ for i in range(self.n_xf_blocks):
467
+ blocks.append(
468
+ (
469
+ f"block_{i}",
470
+ TransformerBlock(self.n_state, 2 * self.n_xf_blocks, attn_fn, self.device),
471
+ )
472
+ )
473
+
474
+ blocks.append(("output", ImageFeatureExtractor(self.n_state, self.n_embd, self.device)))
475
+
476
+ self.blocks = nn.ModuleDict(OrderedDict(blocks))
477
+
478
+ def forward(
479
+ self,
480
+ image: torch.Tensor,
481
+ timesteps: Optional[torch.Tensor] = None,
482
+ return_probe_features: bool = False,
483
+ ) -> torch.Tensor:
484
+ n_batch = image.shape[0]
485
+ h = self.blocks["input"](image, t=timesteps)
486
+
487
+ for i in range(self.n_xf_blocks):
488
+ h = self.blocks[f"block_{i}"](h)
489
+
490
+ h = self.blocks["output"](h, return_probe_features=return_probe_features)
491
+
492
+ assert list(h.shape) == [
493
+ n_batch,
494
+ self.n_embd if not return_probe_features else self.n_state,
495
+ ]
496
+
497
+ return h
glide_text2im/clip/model_creation.py ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ from functools import lru_cache
3
+ from typing import Any, Callable, Dict, List, Optional, Tuple
4
+
5
+ import attr
6
+ import numpy as np
7
+ import torch
8
+ import torch.nn as nn
9
+ import yaml
10
+ from glide_text2im.tokenizer.simple_tokenizer import SimpleTokenizer
11
+
12
+ from .encoders import ImageEncoder, TextEncoder
13
+
14
+
15
+ @lru_cache()
16
+ def default_config_path() -> str:
17
+ return os.path.join(os.path.dirname(os.path.abspath(__file__)), "config.yaml")
18
+
19
+
20
+ @attr.s
21
+ class CLIPModel:
22
+ config: Dict[str, Any] = attr.ib()
23
+ text_encoder: nn.Module = attr.ib()
24
+ image_encoder: nn.Module = attr.ib()
25
+ logit_scale: torch.Tensor = attr.ib()
26
+ device: torch.device = attr.ib()
27
+ tokenizer: SimpleTokenizer = attr.ib()
28
+
29
+ def encode_prompts(self, prompts: List[str]) -> Tuple[torch.Tensor, torch.Tensor]:
30
+ tokens = []
31
+ lens = []
32
+ for prompt in prompts:
33
+ sub_tokens, sub_len = self.tokenizer.padded_tokens_and_len(
34
+ self.tokenizer.encode(prompt), self.text_encoder.max_text_len
35
+ )
36
+ tokens.append(sub_tokens)
37
+ lens.append(sub_len)
38
+ return (
39
+ torch.tensor(tokens).to(dtype=torch.long, device=self.device),
40
+ torch.tensor(lens).to(dtype=torch.long, device=self.device),
41
+ )
42
+
43
+ def text_embeddings(self, prompts: List[str]) -> torch.Tensor:
44
+ tokens, lens = self.encode_prompts(prompts)
45
+ z_t = self.text_encoder(tokens, lens)
46
+ return z_t / (torch.linalg.norm(z_t, dim=-1, keepdim=True) + 1e-12)
47
+
48
+ def image_embeddings(self, images: torch.Tensor, t: torch.Tensor) -> torch.Tensor:
49
+ z_i = self.image_encoder((images + 1) * 127.5, t)
50
+ return z_i / (torch.linalg.norm(z_i, dim=-1, keepdim=True) + 1e-12)
51
+
52
+ def cond_fn(self, prompts: List[str], grad_scale: float) -> Callable[..., torch.Tensor]:
53
+ with torch.no_grad():
54
+ z_t = self.text_embeddings(prompts)
55
+
56
+ def cond_fn(x, t, grad_scale=grad_scale, **kwargs):
57
+ with torch.enable_grad():
58
+ x_var = x.detach().requires_grad_(True)
59
+ z_i = self.image_embeddings(x_var, t)
60
+ loss = torch.exp(self.logit_scale) * (z_t * z_i).sum()
61
+ grad = torch.autograd.grad(loss, x_var)[0].detach()
62
+ return grad * grad_scale
63
+
64
+ return cond_fn
65
+
66
+
67
+ def create_clip_model(
68
+ config_path: Optional[str] = None,
69
+ device: Optional[torch.device] = None,
70
+ tokenizer: Optional[SimpleTokenizer] = None,
71
+ ) -> CLIPModel:
72
+ if config_path is None:
73
+ config_path = default_config_path()
74
+ if device is None:
75
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
76
+ if tokenizer is None:
77
+ tokenizer = SimpleTokenizer()
78
+
79
+ with open(config_path, "r") as f:
80
+ config = yaml.load(f, Loader=yaml.SafeLoader)
81
+
82
+ text_encoder = TextEncoder(
83
+ n_bpe_vocab=config["n_vocab"],
84
+ max_text_len=config["max_text_len"],
85
+ n_embd=config["n_embd"],
86
+ n_head=config["n_head_text"],
87
+ n_xf_blocks=config["n_xf_blocks_text"],
88
+ n_head_state=config["n_head_state_text"],
89
+ device=device,
90
+ )
91
+
92
+ image_encoder = ImageEncoder(
93
+ image_size=config["image_size"],
94
+ patch_size=config["patch_size"],
95
+ n_embd=config["n_embd"],
96
+ n_head=config["n_head_image"],
97
+ n_xf_blocks=config["n_xf_blocks_image"],
98
+ n_head_state=config["n_head_state_image"],
99
+ n_timestep=config["n_timesteps"],
100
+ device=device,
101
+ )
102
+
103
+ logit_scale = torch.tensor(
104
+ np.log(config["logit_scale"]),
105
+ dtype=torch.float32,
106
+ device=device,
107
+ requires_grad=False,
108
+ )
109
+
110
+ return CLIPModel(
111
+ config=config,
112
+ text_encoder=text_encoder,
113
+ image_encoder=image_encoder,
114
+ logit_scale=logit_scale,
115
+ device=device,
116
+ tokenizer=tokenizer,
117
+ )
glide_text2im/clip/utils.py ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import math
2
+ from typing import Callable, Optional
3
+
4
+ import attr
5
+ import torch
6
+ import torch.nn as nn
7
+ import torch.nn.functional as F
8
+
9
+ FilterFn = Callable[[torch.Tensor], torch.Tensor]
10
+
11
+
12
+ class ZeroKeyBiasGrad(torch.autograd.Function):
13
+ @staticmethod
14
+ def forward(ctx, x):
15
+ return x
16
+
17
+ @staticmethod
18
+ def backward(ctx, output_grad):
19
+ output_grad = output_grad.clone()
20
+ output_grad.chunk(3)[1].zero_()
21
+ return output_grad
22
+
23
+
24
+ def zero_key_bias_grad(x: torch.Tensor) -> torch.Tensor:
25
+ return ZeroKeyBiasGrad.apply(x)
26
+
27
+
28
+ @attr.s(eq=False, repr=False)
29
+ class LayerNorm(nn.Module):
30
+ n_state: int = attr.ib()
31
+ eps: float = attr.ib(default=1e-6)
32
+ device: torch.device = attr.ib(default=torch.device("cuda"))
33
+
34
+ def __attrs_post_init__(self) -> None:
35
+ super().__init__()
36
+ self.g = nn.Parameter(torch.ones((self.n_state,), dtype=torch.float32, device=self.device))
37
+ self.b = nn.Parameter(torch.zeros((self.n_state,), dtype=torch.float32, device=self.device))
38
+ self.g.weight_decay_level = "disable" # type: ignore
39
+ self.b.weight_decay_level = "disable" # type: ignore
40
+
41
+ def forward(self, x: torch.Tensor) -> torch.Tensor:
42
+ return F.layer_norm(
43
+ x.type(torch.float32), torch.Size((self.n_state,)), self.g, self.b, self.eps
44
+ )
45
+
46
+
47
+ @attr.s(eq=False, repr=False)
48
+ class Affine(nn.Module):
49
+ n_in: int = attr.ib()
50
+ n_out: int = attr.ib()
51
+ use_bias: bool = attr.ib(default=True)
52
+ use_admnet_init: bool = attr.ib(default=False)
53
+ std: Optional[float] = attr.ib(default=None)
54
+ extra_init_scale: Optional[float] = attr.ib(default=None)
55
+ bias_filter_fn: FilterFn = attr.ib(default=lambda x: x)
56
+ device: torch.device = attr.ib(default=torch.device("cuda"))
57
+
58
+ def __attrs_post_init__(self) -> None:
59
+ super().__init__()
60
+
61
+ if not self.use_admnet_init:
62
+ self.std = self.std if self.std is not None else math.sqrt(2 / (self.n_in + self.n_out))
63
+ self.std = (
64
+ self.std if self.extra_init_scale is None else self.std * self.extra_init_scale
65
+ )
66
+
67
+ w = torch.empty((self.n_out, self.n_in), dtype=torch.float32, device=self.device)
68
+ self.w = nn.Parameter(w)
69
+
70
+ if self.use_bias:
71
+ self.b = nn.Parameter(
72
+ torch.zeros((self.n_out,), dtype=torch.float32, device=self.device)
73
+ )
74
+ self.b.weight_decay_level = "disable" # type: ignore
75
+ else:
76
+ if self.extra_init_scale is not None:
77
+ raise ValueError("extra_init_scale incompatible with admnet init")
78
+
79
+ w = torch.empty((self.n_out, self.n_in), dtype=torch.float32, device=self.device)
80
+
81
+ if self.use_bias:
82
+ b = torch.empty((self.n_out,), dtype=torch.float32, device=self.device)
83
+
84
+ self.w = nn.Parameter(w)
85
+
86
+ if self.use_bias:
87
+ self.b = nn.Parameter(b)
88
+ self.b.weight_decay_level = "disable" # type: ignore
89
+
90
+ def forward(self, x: torch.Tensor) -> torch.Tensor:
91
+ w = self.w if self.w.dtype == x.dtype else self.w.to(x.dtype)
92
+ b = (
93
+ self.bias_filter_fn(self.b if self.b.dtype == x.dtype else self.b.to(x.dtype))
94
+ if self.use_bias
95
+ else None
96
+ )
97
+ return F.linear(x, w, b)
glide_text2im/download.py ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ from functools import lru_cache
3
+ from typing import Dict, Optional
4
+
5
+ import requests
6
+ import torch as th
7
+ from filelock import FileLock
8
+ from tqdm.auto import tqdm
9
+
10
+ MODEL_PATHS = {
11
+ "base": "https://openaipublic.blob.core.windows.net/diffusion/dec-2021/base.pt",
12
+ "upsample": "https://openaipublic.blob.core.windows.net/diffusion/dec-2021/upsample.pt",
13
+ "base-inpaint": "https://openaipublic.blob.core.windows.net/diffusion/dec-2021/base_inpaint.pt",
14
+ "upsample-inpaint": "https://openaipublic.blob.core.windows.net/diffusion/dec-2021/upsample_inpaint.pt",
15
+ "clip/image-enc": "https://openaipublic.blob.core.windows.net/diffusion/dec-2021/clip_image_enc.pt",
16
+ "clip/text-enc": "https://openaipublic.blob.core.windows.net/diffusion/dec-2021/clip_text_enc.pt",
17
+ }
18
+
19
+
20
+ @lru_cache()
21
+ def default_cache_dir() -> str:
22
+ return os.path.join(os.path.abspath(os.getcwd()), "glide_model_cache")
23
+
24
+
25
+ def fetch_file_cached(
26
+ url: str, progress: bool = True, cache_dir: Optional[str] = None, chunk_size: int = 4096
27
+ ) -> str:
28
+ """
29
+ Download the file at the given URL into a local file and return the path.
30
+
31
+ If cache_dir is specified, it will be used to download the files.
32
+ Otherwise, default_cache_dir() is used.
33
+ """
34
+ if cache_dir is None:
35
+ cache_dir = default_cache_dir()
36
+ os.makedirs(cache_dir, exist_ok=True)
37
+ response = requests.get(url, stream=True)
38
+ size = int(response.headers.get("content-length", "0"))
39
+ local_path = os.path.join(cache_dir, url.split("/")[-1])
40
+ with FileLock(local_path + ".lock"):
41
+ if os.path.exists(local_path):
42
+ return local_path
43
+ if progress:
44
+ pbar = tqdm(total=size, unit="iB", unit_scale=True)
45
+ tmp_path = local_path + ".tmp"
46
+ with open(tmp_path, "wb") as f:
47
+ for chunk in response.iter_content(chunk_size):
48
+ if progress:
49
+ pbar.update(len(chunk))
50
+ f.write(chunk)
51
+ os.rename(tmp_path, local_path)
52
+ if progress:
53
+ pbar.close()
54
+ return local_path
55
+
56
+
57
+ def load_checkpoint(
58
+ checkpoint_name: str,
59
+ device: th.device,
60
+ progress: bool = True,
61
+ cache_dir: Optional[str] = None,
62
+ chunk_size: int = 4096,
63
+ ) -> Dict[str, th.Tensor]:
64
+ if checkpoint_name not in MODEL_PATHS:
65
+ raise ValueError(
66
+ f"Unknown checkpoint name {checkpoint_name}. Known names are: {MODEL_PATHS.keys()}."
67
+ )
68
+ path = fetch_file_cached(
69
+ MODEL_PATHS[checkpoint_name], progress=progress, cache_dir=cache_dir, chunk_size=chunk_size
70
+ )
71
+ return th.load(path, map_location=device)
glide_text2im/fp16_util.py ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Helpers to inference with 16-bit precision.
3
+ """
4
+
5
+ import torch.nn as nn
6
+
7
+
8
+ def convert_module_to_f16(l):
9
+ """
10
+ Convert primitive modules to float16.
11
+ """
12
+ if isinstance(l, (nn.Conv1d, nn.Conv2d, nn.Conv3d)):
13
+ l.weight.data = l.weight.data.half()
14
+ if l.bias is not None:
15
+ l.bias.data = l.bias.data.half()
16
+
17
+
18
+ def convert_module_to_f32(l):
19
+ """
20
+ Convert primitive modules to float32, undoing convert_module_to_f16().
21
+ """
22
+ if isinstance(l, (nn.Conv1d, nn.Conv2d, nn.Conv3d)):
23
+ l.weight.data = l.weight.data.float()
24
+ if l.bias is not None:
25
+ l.bias.data = l.bias.data.float()
glide_text2im/gaussian_diffusion.py ADDED
@@ -0,0 +1,639 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Simplified from https://github.com/openai/guided-diffusion/blob/main/guided_diffusion/gaussian_diffusion.py.
3
+ """
4
+
5
+ import math
6
+
7
+ import numpy as np
8
+ import torch as th
9
+
10
+
11
+ def _warmup_beta(beta_start, beta_end, num_diffusion_timesteps, warmup_frac):
12
+ betas = beta_end * np.ones(num_diffusion_timesteps, dtype=np.float64)
13
+ warmup_time = int(num_diffusion_timesteps * warmup_frac)
14
+ betas[:warmup_time] = np.linspace(beta_start, beta_end, warmup_time, dtype=np.float64)
15
+ return betas
16
+
17
+
18
+ def get_beta_schedule(beta_schedule, *, beta_start, beta_end, num_diffusion_timesteps):
19
+ """
20
+ This is the deprecated API for creating beta schedules.
21
+
22
+ See get_named_beta_schedule() for the new library of schedules.
23
+ """
24
+ if beta_schedule == "quad":
25
+ betas = (
26
+ np.linspace(
27
+ beta_start ** 0.5,
28
+ beta_end ** 0.5,
29
+ num_diffusion_timesteps,
30
+ dtype=np.float64,
31
+ )
32
+ ** 2
33
+ )
34
+ elif beta_schedule == "linear":
35
+ betas = np.linspace(beta_start, beta_end, num_diffusion_timesteps, dtype=np.float64)
36
+ elif beta_schedule == "warmup10":
37
+ betas = _warmup_beta(beta_start, beta_end, num_diffusion_timesteps, 0.1)
38
+ elif beta_schedule == "warmup50":
39
+ betas = _warmup_beta(beta_start, beta_end, num_diffusion_timesteps, 0.5)
40
+ elif beta_schedule == "const":
41
+ betas = beta_end * np.ones(num_diffusion_timesteps, dtype=np.float64)
42
+ elif beta_schedule == "jsd": # 1/T, 1/(T-1), 1/(T-2), ..., 1
43
+ betas = 1.0 / np.linspace(
44
+ num_diffusion_timesteps, 1, num_diffusion_timesteps, dtype=np.float64
45
+ )
46
+ else:
47
+ raise NotImplementedError(beta_schedule)
48
+ assert betas.shape == (num_diffusion_timesteps,)
49
+ return betas
50
+
51
+
52
+ def get_named_beta_schedule(schedule_name, num_diffusion_timesteps):
53
+ """
54
+ Get a pre-defined beta schedule for the given name.
55
+
56
+ The beta schedule library consists of beta schedules which remain similar
57
+ in the limit of num_diffusion_timesteps.
58
+ Beta schedules may be added, but should not be removed or changed once
59
+ they are committed to maintain backwards compatibility.
60
+ """
61
+ if schedule_name == "linear":
62
+ # Linear schedule from Ho et al, extended to work for any number of
63
+ # diffusion steps.
64
+ scale = 1000 / num_diffusion_timesteps
65
+ return get_beta_schedule(
66
+ "linear",
67
+ beta_start=scale * 0.0001,
68
+ beta_end=scale * 0.02,
69
+ num_diffusion_timesteps=num_diffusion_timesteps,
70
+ )
71
+ elif schedule_name == "squaredcos_cap_v2":
72
+ return betas_for_alpha_bar(
73
+ num_diffusion_timesteps,
74
+ lambda t: math.cos((t + 0.008) / 1.008 * math.pi / 2) ** 2,
75
+ )
76
+ else:
77
+ raise NotImplementedError(f"unknown beta schedule: {schedule_name}")
78
+
79
+
80
+ def betas_for_alpha_bar(num_diffusion_timesteps, alpha_bar, max_beta=0.999):
81
+ """
82
+ Create a beta schedule that discretizes the given alpha_t_bar function,
83
+ which defines the cumulative product of (1-beta) over time from t = [0,1].
84
+
85
+ :param num_diffusion_timesteps: the number of betas to produce.
86
+ :param alpha_bar: a lambda that takes an argument t from 0 to 1 and
87
+ produces the cumulative product of (1-beta) up to that
88
+ part of the diffusion process.
89
+ :param max_beta: the maximum beta to use; use values lower than 1 to
90
+ prevent singularities.
91
+ """
92
+ betas = []
93
+ for i in range(num_diffusion_timesteps):
94
+ t1 = i / num_diffusion_timesteps
95
+ t2 = (i + 1) / num_diffusion_timesteps
96
+ betas.append(min(1 - alpha_bar(t2) / alpha_bar(t1), max_beta))
97
+ return np.array(betas)
98
+
99
+
100
+ class GaussianDiffusion:
101
+ """
102
+ Utilities for training and sampling diffusion models.
103
+
104
+ Original ported from this codebase:
105
+ https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/diffusion_utils_2.py#L42
106
+
107
+ :param betas: a 1-D numpy array of betas for each diffusion timestep,
108
+ starting at T and going to 1.
109
+ """
110
+
111
+ def __init__(
112
+ self,
113
+ *,
114
+ betas,
115
+ ):
116
+ # Use float64 for accuracy.
117
+ betas = np.array(betas, dtype=np.float64)
118
+ self.betas = betas
119
+ assert len(betas.shape) == 1, "betas must be 1-D"
120
+ assert (betas > 0).all() and (betas <= 1).all()
121
+
122
+ self.num_timesteps = int(betas.shape[0])
123
+
124
+ alphas = 1.0 - betas
125
+ self.alphas_cumprod = np.cumprod(alphas, axis=0)
126
+ self.alphas_cumprod_prev = np.append(1.0, self.alphas_cumprod[:-1])
127
+ self.alphas_cumprod_next = np.append(self.alphas_cumprod[1:], 0.0)
128
+ assert self.alphas_cumprod_prev.shape == (self.num_timesteps,)
129
+
130
+ # calculations for diffusion q(x_t | x_{t-1}) and others
131
+ self.sqrt_alphas_cumprod = np.sqrt(self.alphas_cumprod)
132
+ self.sqrt_one_minus_alphas_cumprod = np.sqrt(1.0 - self.alphas_cumprod)
133
+ self.log_one_minus_alphas_cumprod = np.log(1.0 - self.alphas_cumprod)
134
+ self.sqrt_recip_alphas_cumprod = np.sqrt(1.0 / self.alphas_cumprod)
135
+ self.sqrt_recipm1_alphas_cumprod = np.sqrt(1.0 / self.alphas_cumprod - 1)
136
+
137
+ # calculations for posterior q(x_{t-1} | x_t, x_0)
138
+ self.posterior_variance = (
139
+ betas * (1.0 - self.alphas_cumprod_prev) / (1.0 - self.alphas_cumprod)
140
+ )
141
+ # below: log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain
142
+ self.posterior_log_variance_clipped = np.log(
143
+ np.append(self.posterior_variance[1], self.posterior_variance[1:])
144
+ )
145
+ self.posterior_mean_coef1 = (
146
+ betas * np.sqrt(self.alphas_cumprod_prev) / (1.0 - self.alphas_cumprod)
147
+ )
148
+ self.posterior_mean_coef2 = (
149
+ (1.0 - self.alphas_cumprod_prev) * np.sqrt(alphas) / (1.0 - self.alphas_cumprod)
150
+ )
151
+
152
+ def q_mean_variance(self, x_start, t):
153
+ """
154
+ Get the distribution q(x_t | x_0).
155
+
156
+ :param x_start: the [N x C x ...] tensor of noiseless inputs.
157
+ :param t: the number of diffusion steps (minus 1). Here, 0 means one step.
158
+ :return: A tuple (mean, variance, log_variance), all of x_start's shape.
159
+ """
160
+ mean = _extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
161
+ variance = _extract_into_tensor(1.0 - self.alphas_cumprod, t, x_start.shape)
162
+ log_variance = _extract_into_tensor(self.log_one_minus_alphas_cumprod, t, x_start.shape)
163
+ return mean, variance, log_variance
164
+
165
+ def q_sample(self, x_start, t, noise=None):
166
+ """
167
+ Diffuse the data for a given number of diffusion steps.
168
+
169
+ In other words, sample from q(x_t | x_0).
170
+
171
+ :param x_start: the initial data batch.
172
+ :param t: the number of diffusion steps (minus 1). Here, 0 means one step.
173
+ :param noise: if specified, the split-out normal noise.
174
+ :return: A noisy version of x_start.
175
+ """
176
+ if noise is None:
177
+ noise = th.randn_like(x_start)
178
+ assert noise.shape == x_start.shape
179
+ return (
180
+ _extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
181
+ + _extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
182
+ )
183
+
184
+ def q_posterior_mean_variance(self, x_start, x_t, t):
185
+ """
186
+ Compute the mean and variance of the diffusion posterior:
187
+
188
+ q(x_{t-1} | x_t, x_0)
189
+
190
+ """
191
+ assert x_start.shape == x_t.shape
192
+ posterior_mean = (
193
+ _extract_into_tensor(self.posterior_mean_coef1, t, x_t.shape) * x_start
194
+ + _extract_into_tensor(self.posterior_mean_coef2, t, x_t.shape) * x_t
195
+ )
196
+ posterior_variance = _extract_into_tensor(self.posterior_variance, t, x_t.shape)
197
+ posterior_log_variance_clipped = _extract_into_tensor(
198
+ self.posterior_log_variance_clipped, t, x_t.shape
199
+ )
200
+ assert (
201
+ posterior_mean.shape[0]
202
+ == posterior_variance.shape[0]
203
+ == posterior_log_variance_clipped.shape[0]
204
+ == x_start.shape[0]
205
+ )
206
+ return posterior_mean, posterior_variance, posterior_log_variance_clipped
207
+
208
+ def p_mean_variance(self, model, x, t, clip_denoised=True, denoised_fn=None, model_kwargs=None):
209
+ """
210
+ Apply the model to get p(x_{t-1} | x_t), as well as a prediction of
211
+ the initial x, x_0.
212
+
213
+ :param model: the model, which takes a signal and a batch of timesteps
214
+ as input.
215
+ :param x: the [N x C x ...] tensor at time t.
216
+ :param t: a 1-D Tensor of timesteps.
217
+ :param clip_denoised: if True, clip the denoised signal into [-1, 1].
218
+ :param denoised_fn: if not None, a function which applies to the
219
+ x_start prediction before it is used to sample. Applies before
220
+ clip_denoised.
221
+ :param model_kwargs: if not None, a dict of extra keyword arguments to
222
+ pass to the model. This can be used for conditioning.
223
+ :return: a dict with the following keys:
224
+ - 'mean': the model mean output.
225
+ - 'variance': the model variance output.
226
+ - 'log_variance': the log of 'variance'.
227
+ - 'pred_xstart': the prediction for x_0.
228
+ """
229
+ if model_kwargs is None:
230
+ model_kwargs = {}
231
+
232
+ B, C = x.shape[:2]
233
+ assert t.shape == (B,)
234
+ model_output = model(x, t, **model_kwargs)
235
+ if isinstance(model_output, tuple):
236
+ model_output, extra = model_output
237
+ else:
238
+ extra = None
239
+
240
+ assert model_output.shape == (B, C * 2, *x.shape[2:])
241
+ model_output, model_var_values = th.split(model_output, C, dim=1)
242
+ min_log = _extract_into_tensor(self.posterior_log_variance_clipped, t, x.shape)
243
+ max_log = _extract_into_tensor(np.log(self.betas), t, x.shape)
244
+ # The model_var_values is [-1, 1] for [min_var, max_var].
245
+ frac = (model_var_values + 1) / 2
246
+ model_log_variance = frac * max_log + (1 - frac) * min_log
247
+ model_variance = th.exp(model_log_variance)
248
+
249
+ def process_xstart(x):
250
+ if denoised_fn is not None:
251
+ x = denoised_fn(x)
252
+ if clip_denoised:
253
+ return x.clamp(-1, 1)
254
+ return x
255
+
256
+ pred_xstart = process_xstart(self._predict_xstart_from_eps(x_t=x, t=t, eps=model_output))
257
+ model_mean, _, _ = self.q_posterior_mean_variance(x_start=pred_xstart, x_t=x, t=t)
258
+
259
+ assert model_mean.shape == model_log_variance.shape == pred_xstart.shape == x.shape
260
+ return {
261
+ "mean": model_mean,
262
+ "variance": model_variance,
263
+ "log_variance": model_log_variance,
264
+ "pred_xstart": pred_xstart,
265
+ "extra": extra,
266
+ }
267
+
268
+ def _predict_xstart_from_eps(self, x_t, t, eps):
269
+ assert x_t.shape == eps.shape
270
+ return (
271
+ _extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t
272
+ - _extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * eps
273
+ )
274
+
275
+ def _predict_eps_from_xstart(self, x_t, t, pred_xstart):
276
+ return (
277
+ _extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - pred_xstart
278
+ ) / _extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape)
279
+
280
+ def condition_mean(self, cond_fn, p_mean_var, x, t, model_kwargs=None):
281
+ """
282
+ Compute the mean for the previous step, given a function cond_fn that
283
+ computes the gradient of a conditional log probability with respect to
284
+ x. In particular, cond_fn computes grad(log(p(y|x))), and we want to
285
+ condition on y.
286
+
287
+ This uses the conditioning strategy from Sohl-Dickstein et al. (2015).
288
+ """
289
+ gradient = cond_fn(x, t, **model_kwargs)
290
+ new_mean = p_mean_var["mean"].float() + p_mean_var["variance"] * gradient.float()
291
+ return new_mean
292
+
293
+ def condition_score(self, cond_fn, p_mean_var, x, t, model_kwargs=None):
294
+ """
295
+ Compute what the p_mean_variance output would have been, should the
296
+ model's score function be conditioned by cond_fn.
297
+
298
+ See condition_mean() for details on cond_fn.
299
+
300
+ Unlike condition_mean(), this instead uses the conditioning strategy
301
+ from Song et al (2020).
302
+ """
303
+ alpha_bar = _extract_into_tensor(self.alphas_cumprod, t, x.shape)
304
+
305
+ eps = self._predict_eps_from_xstart(x, t, p_mean_var["pred_xstart"])
306
+ eps = eps - (1 - alpha_bar).sqrt() * cond_fn(x, t, **model_kwargs)
307
+
308
+ out = p_mean_var.copy()
309
+ out["pred_xstart"] = self._predict_xstart_from_eps(x, t, eps)
310
+ out["mean"], _, _ = self.q_posterior_mean_variance(x_start=out["pred_xstart"], x_t=x, t=t)
311
+ return out
312
+
313
+ def p_sample(
314
+ self,
315
+ model,
316
+ x,
317
+ t,
318
+ clip_denoised=True,
319
+ denoised_fn=None,
320
+ cond_fn=None,
321
+ model_kwargs=None,
322
+ ):
323
+ """
324
+ Sample x_{t-1} from the model at the given timestep.
325
+
326
+ :param model: the model to sample from.
327
+ :param x: the current tensor at x_{t-1}.
328
+ :param t: the value of t, starting at 0 for the first diffusion step.
329
+ :param clip_denoised: if True, clip the x_start prediction to [-1, 1].
330
+ :param denoised_fn: if not None, a function which applies to the
331
+ x_start prediction before it is used to sample.
332
+ :param cond_fn: if not None, this is a gradient function that acts
333
+ similarly to the model.
334
+ :param model_kwargs: if not None, a dict of extra keyword arguments to
335
+ pass to the model. This can be used for conditioning.
336
+ :return: a dict containing the following keys:
337
+ - 'sample': a random sample from the model.
338
+ - 'pred_xstart': a prediction of x_0.
339
+ """
340
+ out = self.p_mean_variance(
341
+ model,
342
+ x,
343
+ t,
344
+ clip_denoised=clip_denoised,
345
+ denoised_fn=denoised_fn,
346
+ model_kwargs=model_kwargs,
347
+ )
348
+ noise = th.randn_like(x)
349
+ nonzero_mask = (
350
+ (t != 0).float().view(-1, *([1] * (len(x.shape) - 1)))
351
+ ) # no noise when t == 0
352
+ if cond_fn is not None:
353
+ out["mean"] = self.condition_mean(cond_fn, out, x, t, model_kwargs=model_kwargs)
354
+ sample = out["mean"] + nonzero_mask * th.exp(0.5 * out["log_variance"]) * noise
355
+ return {"sample": sample, "pred_xstart": out["pred_xstart"]}
356
+
357
+ def p_sample_loop(
358
+ self,
359
+ model,
360
+ shape,
361
+ noise=None,
362
+ clip_denoised=True,
363
+ denoised_fn=None,
364
+ cond_fn=None,
365
+ model_kwargs=None,
366
+ device=None,
367
+ progress=False,
368
+ ):
369
+ """
370
+ Generate samples from the model.
371
+
372
+ :param model: the model module.
373
+ :param shape: the shape of the samples, (N, C, H, W).
374
+ :param noise: if specified, the noise from the encoder to sample.
375
+ Should be of the same shape as `shape`.
376
+ :param clip_denoised: if True, clip x_start predictions to [-1, 1].
377
+ :param denoised_fn: if not None, a function which applies to the
378
+ x_start prediction before it is used to sample.
379
+ :param cond_fn: if not None, this is a gradient function that acts
380
+ similarly to the model.
381
+ :param model_kwargs: if not None, a dict of extra keyword arguments to
382
+ pass to the model. This can be used for conditioning.
383
+ :param device: if specified, the device to create the samples on.
384
+ If not specified, use a model parameter's device.
385
+ :param progress: if True, show a tqdm progress bar.
386
+ :return: a non-differentiable batch of samples.
387
+ """
388
+ final = None
389
+ for sample in self.p_sample_loop_progressive(
390
+ model,
391
+ shape,
392
+ noise=noise,
393
+ clip_denoised=clip_denoised,
394
+ denoised_fn=denoised_fn,
395
+ cond_fn=cond_fn,
396
+ model_kwargs=model_kwargs,
397
+ device=device,
398
+ progress=progress,
399
+ ):
400
+ final = sample
401
+ return final["sample"]
402
+
403
+ def p_sample_loop_progressive(
404
+ self,
405
+ model,
406
+ shape,
407
+ noise=None,
408
+ clip_denoised=True,
409
+ denoised_fn=None,
410
+ cond_fn=None,
411
+ model_kwargs=None,
412
+ device=None,
413
+ progress=False,
414
+ ):
415
+ """
416
+ Generate samples from the model and yield intermediate samples from
417
+ each timestep of diffusion.
418
+
419
+ Arguments are the same as p_sample_loop().
420
+ Returns a generator over dicts, where each dict is the return value of
421
+ p_sample().
422
+ """
423
+ if device is None:
424
+ device = next(model.parameters()).device
425
+ assert isinstance(shape, (tuple, list))
426
+ if noise is not None:
427
+ img = noise
428
+ else:
429
+ img = th.randn(*shape, device=device)
430
+ indices = list(range(self.num_timesteps))[::-1]
431
+
432
+ if progress:
433
+ # Lazy import so that we don't depend on tqdm.
434
+ from tqdm.auto import tqdm
435
+
436
+ indices = tqdm(indices)
437
+
438
+ for i in indices:
439
+ t = th.tensor([i] * shape[0], device=device)
440
+ with th.no_grad():
441
+ out = self.p_sample(
442
+ model,
443
+ img,
444
+ t,
445
+ clip_denoised=clip_denoised,
446
+ denoised_fn=denoised_fn,
447
+ cond_fn=cond_fn,
448
+ model_kwargs=model_kwargs,
449
+ )
450
+ yield out
451
+ img = out["sample"]
452
+
453
+ def ddim_sample(
454
+ self,
455
+ model,
456
+ x,
457
+ t,
458
+ clip_denoised=True,
459
+ denoised_fn=None,
460
+ cond_fn=None,
461
+ model_kwargs=None,
462
+ eta=0.0,
463
+ ):
464
+ """
465
+ Sample x_{t-1} from the model using DDIM.
466
+
467
+ Same usage as p_sample().
468
+ """
469
+ out = self.p_mean_variance(
470
+ model,
471
+ x,
472
+ t,
473
+ clip_denoised=clip_denoised,
474
+ denoised_fn=denoised_fn,
475
+ model_kwargs=model_kwargs,
476
+ )
477
+ if cond_fn is not None:
478
+ out = self.condition_score(cond_fn, out, x, t, model_kwargs=model_kwargs)
479
+
480
+ # Usually our model outputs epsilon, but we re-derive it
481
+ # in case we used x_start or x_prev prediction.
482
+ eps = self._predict_eps_from_xstart(x, t, out["pred_xstart"])
483
+
484
+ alpha_bar = _extract_into_tensor(self.alphas_cumprod, t, x.shape)
485
+ alpha_bar_prev = _extract_into_tensor(self.alphas_cumprod_prev, t, x.shape)
486
+ sigma = (
487
+ eta
488
+ * th.sqrt((1 - alpha_bar_prev) / (1 - alpha_bar))
489
+ * th.sqrt(1 - alpha_bar / alpha_bar_prev)
490
+ )
491
+ # Equation 12.
492
+ noise = th.randn_like(x)
493
+ mean_pred = (
494
+ out["pred_xstart"] * th.sqrt(alpha_bar_prev)
495
+ + th.sqrt(1 - alpha_bar_prev - sigma ** 2) * eps
496
+ )
497
+ nonzero_mask = (
498
+ (t != 0).float().view(-1, *([1] * (len(x.shape) - 1)))
499
+ ) # no noise when t == 0
500
+ sample = mean_pred + nonzero_mask * sigma * noise
501
+ return {"sample": sample, "pred_xstart": out["pred_xstart"]}
502
+
503
+ def ddim_reverse_sample(
504
+ self,
505
+ model,
506
+ x,
507
+ t,
508
+ clip_denoised=True,
509
+ denoised_fn=None,
510
+ cond_fn=None,
511
+ model_kwargs=None,
512
+ eta=0.0,
513
+ ):
514
+ """
515
+ Sample x_{t+1} from the model using DDIM reverse ODE.
516
+ """
517
+ assert eta == 0.0, "Reverse ODE only for deterministic path"
518
+ out = self.p_mean_variance(
519
+ model,
520
+ x,
521
+ t,
522
+ clip_denoised=clip_denoised,
523
+ denoised_fn=denoised_fn,
524
+ model_kwargs=model_kwargs,
525
+ )
526
+ if cond_fn is not None:
527
+ out = self.condition_score(cond_fn, out, x, t, model_kwargs=model_kwargs)
528
+ # Usually our model outputs epsilon, but we re-derive it
529
+ # in case we used x_start or x_prev prediction.
530
+ eps = (
531
+ _extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x.shape) * x
532
+ - out["pred_xstart"]
533
+ ) / _extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x.shape)
534
+ alpha_bar_next = _extract_into_tensor(self.alphas_cumprod_next, t, x.shape)
535
+
536
+ # Equation 12. reversed
537
+ mean_pred = out["pred_xstart"] * th.sqrt(alpha_bar_next) + th.sqrt(1 - alpha_bar_next) * eps
538
+
539
+ return {"sample": mean_pred, "pred_xstart": out["pred_xstart"]}
540
+
541
+ def ddim_sample_loop(
542
+ self,
543
+ model,
544
+ shape,
545
+ noise=None,
546
+ clip_denoised=True,
547
+ denoised_fn=None,
548
+ cond_fn=None,
549
+ model_kwargs=None,
550
+ device=None,
551
+ progress=False,
552
+ eta=0.0,
553
+ ):
554
+ """
555
+ Generate samples from the model using DDIM.
556
+
557
+ Same usage as p_sample_loop().
558
+ """
559
+ final = None
560
+ for sample in self.ddim_sample_loop_progressive(
561
+ model,
562
+ shape,
563
+ noise=noise,
564
+ clip_denoised=clip_denoised,
565
+ denoised_fn=denoised_fn,
566
+ cond_fn=cond_fn,
567
+ model_kwargs=model_kwargs,
568
+ device=device,
569
+ progress=progress,
570
+ eta=eta,
571
+ ):
572
+ final = sample
573
+ return final["sample"]
574
+
575
+ def ddim_sample_loop_progressive(
576
+ self,
577
+ model,
578
+ shape,
579
+ noise=None,
580
+ clip_denoised=True,
581
+ denoised_fn=None,
582
+ cond_fn=None,
583
+ model_kwargs=None,
584
+ device=None,
585
+ progress=False,
586
+ eta=0.0,
587
+ ):
588
+ """
589
+ Use DDIM to sample from the model and yield intermediate samples from
590
+ each timestep of DDIM.
591
+
592
+ Same usage as p_sample_loop_progressive().
593
+ """
594
+ if device is None:
595
+ device = next(model.parameters()).device
596
+ assert isinstance(shape, (tuple, list))
597
+ if noise is not None:
598
+ img = noise
599
+ else:
600
+ img = th.randn(*shape, device=device)
601
+ indices = list(range(self.num_timesteps))[::-1]
602
+
603
+ if progress:
604
+ # Lazy import so that we don't depend on tqdm.
605
+ from tqdm.auto import tqdm
606
+
607
+ indices = tqdm(indices)
608
+
609
+ for i in indices:
610
+ t = th.tensor([i] * shape[0], device=device)
611
+ with th.no_grad():
612
+ out = self.ddim_sample(
613
+ model,
614
+ img,
615
+ t,
616
+ clip_denoised=clip_denoised,
617
+ denoised_fn=denoised_fn,
618
+ cond_fn=cond_fn,
619
+ model_kwargs=model_kwargs,
620
+ eta=eta,
621
+ )
622
+ yield out
623
+ img = out["sample"]
624
+
625
+
626
+ def _extract_into_tensor(arr, timesteps, broadcast_shape):
627
+ """
628
+ Extract values from a 1-D numpy array for a batch of indices.
629
+
630
+ :param arr: the 1-D numpy array.
631
+ :param timesteps: a tensor of indices into the array to extract.
632
+ :param broadcast_shape: a larger shape of K dimensions with the batch
633
+ dimension equal to the length of timesteps.
634
+ :return: a tensor of shape [batch_size, 1, ...] where the shape has K dims.
635
+ """
636
+ res = th.from_numpy(arr).to(device=timesteps.device)[timesteps].float()
637
+ while len(res.shape) < len(broadcast_shape):
638
+ res = res[..., None]
639
+ return res + th.zeros(broadcast_shape, device=timesteps.device)
glide_text2im/model_creation.py ADDED
@@ -0,0 +1,195 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from glide_text2im.gaussian_diffusion import get_named_beta_schedule
2
+ from glide_text2im.respace import SpacedDiffusion, space_timesteps
3
+ from glide_text2im.text2im_model import (
4
+ InpaintText2ImUNet,
5
+ SuperResInpaintText2ImUnet,
6
+ SuperResText2ImUNet,
7
+ Text2ImUNet,
8
+ )
9
+ from glide_text2im.tokenizer.bpe import get_encoder
10
+
11
+
12
+ def model_and_diffusion_defaults():
13
+ return dict(
14
+ image_size=64,
15
+ num_channels=192,
16
+ num_res_blocks=3,
17
+ channel_mult="",
18
+ num_heads=1,
19
+ num_head_channels=64,
20
+ num_heads_upsample=-1,
21
+ attention_resolutions="32,16,8",
22
+ dropout=0.1,
23
+ text_ctx=128,
24
+ xf_width=512,
25
+ xf_layers=16,
26
+ xf_heads=8,
27
+ xf_final_ln=True,
28
+ xf_padding=True,
29
+ diffusion_steps=1000,
30
+ noise_schedule="squaredcos_cap_v2",
31
+ timestep_respacing="",
32
+ use_scale_shift_norm=True,
33
+ resblock_updown=True,
34
+ use_fp16=True,
35
+ cache_text_emb=False,
36
+ inpaint=False,
37
+ super_res=False,
38
+ )
39
+
40
+
41
+ def model_and_diffusion_defaults_upsampler():
42
+ result = model_and_diffusion_defaults()
43
+ result.update(
44
+ dict(
45
+ image_size=256,
46
+ num_res_blocks=2,
47
+ noise_schedule="linear",
48
+ super_res=True,
49
+ )
50
+ )
51
+ return result
52
+
53
+
54
+ def create_model_and_diffusion(
55
+ image_size,
56
+ num_channels,
57
+ num_res_blocks,
58
+ channel_mult,
59
+ num_heads,
60
+ num_head_channels,
61
+ num_heads_upsample,
62
+ attention_resolutions,
63
+ dropout,
64
+ text_ctx,
65
+ xf_width,
66
+ xf_layers,
67
+ xf_heads,
68
+ xf_final_ln,
69
+ xf_padding,
70
+ diffusion_steps,
71
+ noise_schedule,
72
+ timestep_respacing,
73
+ use_scale_shift_norm,
74
+ resblock_updown,
75
+ use_fp16,
76
+ cache_text_emb,
77
+ inpaint,
78
+ super_res,
79
+ ):
80
+ model = create_model(
81
+ image_size,
82
+ num_channels,
83
+ num_res_blocks,
84
+ channel_mult=channel_mult,
85
+ attention_resolutions=attention_resolutions,
86
+ num_heads=num_heads,
87
+ num_head_channels=num_head_channels,
88
+ num_heads_upsample=num_heads_upsample,
89
+ use_scale_shift_norm=use_scale_shift_norm,
90
+ dropout=dropout,
91
+ text_ctx=text_ctx,
92
+ xf_width=xf_width,
93
+ xf_layers=xf_layers,
94
+ xf_heads=xf_heads,
95
+ xf_final_ln=xf_final_ln,
96
+ xf_padding=xf_padding,
97
+ resblock_updown=resblock_updown,
98
+ use_fp16=use_fp16,
99
+ cache_text_emb=cache_text_emb,
100
+ inpaint=inpaint,
101
+ super_res=super_res,
102
+ )
103
+ diffusion = create_gaussian_diffusion(
104
+ steps=diffusion_steps,
105
+ noise_schedule=noise_schedule,
106
+ timestep_respacing=timestep_respacing,
107
+ )
108
+ return model, diffusion
109
+
110
+
111
+ def create_model(
112
+ image_size,
113
+ num_channels,
114
+ num_res_blocks,
115
+ channel_mult,
116
+ attention_resolutions,
117
+ num_heads,
118
+ num_head_channels,
119
+ num_heads_upsample,
120
+ use_scale_shift_norm,
121
+ dropout,
122
+ text_ctx,
123
+ xf_width,
124
+ xf_layers,
125
+ xf_heads,
126
+ xf_final_ln,
127
+ xf_padding,
128
+ resblock_updown,
129
+ use_fp16,
130
+ cache_text_emb,
131
+ inpaint,
132
+ super_res,
133
+ ):
134
+ if channel_mult == "":
135
+ if image_size == 256:
136
+ channel_mult = (1, 1, 2, 2, 4, 4)
137
+ elif image_size == 128:
138
+ channel_mult = (1, 1, 2, 3, 4)
139
+ elif image_size == 64:
140
+ channel_mult = (1, 2, 3, 4)
141
+ else:
142
+ raise ValueError(f"unsupported image size: {image_size}")
143
+ else:
144
+ channel_mult = tuple(int(ch_mult) for ch_mult in channel_mult.split(","))
145
+ assert 2 ** (len(channel_mult) + 2) == image_size
146
+
147
+ attention_ds = []
148
+ for res in attention_resolutions.split(","):
149
+ attention_ds.append(image_size // int(res))
150
+
151
+ if inpaint and super_res:
152
+ model_cls = SuperResInpaintText2ImUnet
153
+ elif inpaint:
154
+ model_cls = InpaintText2ImUNet
155
+ elif super_res:
156
+ model_cls = SuperResText2ImUNet
157
+ else:
158
+ model_cls = Text2ImUNet
159
+ return model_cls(
160
+ text_ctx=text_ctx,
161
+ xf_width=xf_width,
162
+ xf_layers=xf_layers,
163
+ xf_heads=xf_heads,
164
+ xf_final_ln=xf_final_ln,
165
+ tokenizer=get_encoder(),
166
+ xf_padding=xf_padding,
167
+ in_channels=3,
168
+ model_channels=num_channels,
169
+ out_channels=6,
170
+ num_res_blocks=num_res_blocks,
171
+ attention_resolutions=tuple(attention_ds),
172
+ dropout=dropout,
173
+ channel_mult=channel_mult,
174
+ use_fp16=use_fp16,
175
+ num_heads=num_heads,
176
+ num_head_channels=num_head_channels,
177
+ num_heads_upsample=num_heads_upsample,
178
+ use_scale_shift_norm=use_scale_shift_norm,
179
+ resblock_updown=resblock_updown,
180
+ cache_text_emb=cache_text_emb,
181
+ )
182
+
183
+
184
+ def create_gaussian_diffusion(
185
+ steps,
186
+ noise_schedule,
187
+ timestep_respacing,
188
+ ):
189
+ betas = get_named_beta_schedule(noise_schedule, steps)
190
+ if not timestep_respacing:
191
+ timestep_respacing = [steps]
192
+ return SpacedDiffusion(
193
+ use_timesteps=space_timesteps(steps, timestep_respacing),
194
+ betas=betas,
195
+ )
glide_text2im/nn.py ADDED
@@ -0,0 +1,105 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Various utilities for neural networks.
3
+ """
4
+
5
+ import math
6
+
7
+ import torch as th
8
+ import torch.nn as nn
9
+ import torch.nn.functional as F
10
+
11
+
12
+ class GroupNorm32(nn.GroupNorm):
13
+ def __init__(self, num_groups, num_channels, swish, eps=1e-5):
14
+ super().__init__(num_groups=num_groups, num_channels=num_channels, eps=eps)
15
+ self.swish = swish
16
+
17
+ def forward(self, x):
18
+ y = super().forward(x.float()).to(x.dtype)
19
+ if self.swish == 1.0:
20
+ y = F.silu(y)
21
+ elif self.swish:
22
+ y = y * F.sigmoid(y * float(self.swish))
23
+ return y
24
+
25
+
26
+ def conv_nd(dims, *args, **kwargs):
27
+ """
28
+ Create a 1D, 2D, or 3D convolution module.
29
+ """
30
+ if dims == 1:
31
+ return nn.Conv1d(*args, **kwargs)
32
+ elif dims == 2:
33
+ return nn.Conv2d(*args, **kwargs)
34
+ elif dims == 3:
35
+ return nn.Conv3d(*args, **kwargs)
36
+ raise ValueError(f"unsupported dimensions: {dims}")
37
+
38
+
39
+ def linear(*args, **kwargs):
40
+ """
41
+ Create a linear module.
42
+ """
43
+ return nn.Linear(*args, **kwargs)
44
+
45
+
46
+ def avg_pool_nd(dims, *args, **kwargs):
47
+ """
48
+ Create a 1D, 2D, or 3D average pooling module.
49
+ """
50
+ if dims == 1:
51
+ return nn.AvgPool1d(*args, **kwargs)
52
+ elif dims == 2:
53
+ return nn.AvgPool2d(*args, **kwargs)
54
+ elif dims == 3:
55
+ return nn.AvgPool3d(*args, **kwargs)
56
+ raise ValueError(f"unsupported dimensions: {dims}")
57
+
58
+
59
+ def zero_module(module):
60
+ """
61
+ Zero out the parameters of a module and return it.
62
+ """
63
+ for p in module.parameters():
64
+ p.detach().zero_()
65
+ return module
66
+
67
+
68
+ def scale_module(module, scale):
69
+ """
70
+ Scale the parameters of a module and return it.
71
+ """
72
+ for p in module.parameters():
73
+ p.detach().mul_(scale)
74
+ return module
75
+
76
+
77
+ def normalization(channels, swish=0.0):
78
+ """
79
+ Make a standard normalization layer, with an optional swish activation.
80
+
81
+ :param channels: number of input channels.
82
+ :return: an nn.Module for normalization.
83
+ """
84
+ return GroupNorm32(num_channels=channels, num_groups=32, swish=swish)
85
+
86
+
87
+ def timestep_embedding(timesteps, dim, max_period=10000):
88
+ """
89
+ Create sinusoidal timestep embeddings.
90
+
91
+ :param timesteps: a 1-D Tensor of N indices, one per batch element.
92
+ These may be fractional.
93
+ :param dim: the dimension of the output.
94
+ :param max_period: controls the minimum frequency of the embeddings.
95
+ :return: an [N x dim] Tensor of positional embeddings.
96
+ """
97
+ half = dim // 2
98
+ freqs = th.exp(
99
+ -math.log(max_period) * th.arange(start=0, end=half, dtype=th.float32) / half
100
+ ).to(device=timesteps.device)
101
+ args = timesteps[:, None].float() * freqs[None]
102
+ embedding = th.cat([th.cos(args), th.sin(args)], dim=-1)
103
+ if dim % 2:
104
+ embedding = th.cat([embedding, th.zeros_like(embedding[:, :1])], dim=-1)
105
+ return embedding
glide_text2im/respace.py ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Utilities for changing sampling schedules of a trained model.
3
+
4
+ Simplified from: https://github.com/openai/guided-diffusion/blob/main/guided_diffusion/respace.py
5
+ """
6
+
7
+ import numpy as np
8
+ import torch as th
9
+
10
+ from .gaussian_diffusion import GaussianDiffusion
11
+
12
+
13
+ def space_timesteps(num_timesteps, section_counts):
14
+ """
15
+ Create a list of timesteps to use from an original diffusion process,
16
+ given the number of timesteps we want to take from equally-sized portions
17
+ of the original process.
18
+
19
+ For example, if there's 300 timesteps and the section counts are [10,15,20]
20
+ then the first 100 timesteps are strided to be 10 timesteps, the second 100
21
+ are strided to be 15 timesteps, and the final 100 are strided to be 20.
22
+
23
+ :param num_timesteps: the number of diffusion steps in the original
24
+ process to divide up.
25
+ :param section_counts: either a list of numbers, or a string containing
26
+ comma-separated numbers, indicating the step count
27
+ per section. As a special case, use "ddimN" where N
28
+ is a number of steps to use the striding from the
29
+ DDIM paper.
30
+ :return: a set of diffusion steps from the original process to use.
31
+ """
32
+ if isinstance(section_counts, str):
33
+ if section_counts.startswith("ddim"):
34
+ desired_count = int(section_counts[len("ddim") :])
35
+ for i in range(1, num_timesteps):
36
+ if len(range(0, num_timesteps, i)) == desired_count:
37
+ return set(range(0, num_timesteps, i))
38
+ raise ValueError(f"cannot create exactly {num_timesteps} steps with an integer stride")
39
+ elif section_counts == "fast27":
40
+ steps = space_timesteps(num_timesteps, "10,10,3,2,2")
41
+ # Help reduce DDIM artifacts from noisiest timesteps.
42
+ steps.remove(num_timesteps - 1)
43
+ steps.add(num_timesteps - 3)
44
+ return steps
45
+ section_counts = [int(x) for x in section_counts.split(",")]
46
+ size_per = num_timesteps // len(section_counts)
47
+ extra = num_timesteps % len(section_counts)
48
+ start_idx = 0
49
+ all_steps = []
50
+ for i, section_count in enumerate(section_counts):
51
+ size = size_per + (1 if i < extra else 0)
52
+ if size < section_count:
53
+ raise ValueError(f"cannot divide section of {size} steps into {section_count}")
54
+ if section_count <= 1:
55
+ frac_stride = 1
56
+ else:
57
+ frac_stride = (size - 1) / (section_count - 1)
58
+ cur_idx = 0.0
59
+ taken_steps = []
60
+ for _ in range(section_count):
61
+ taken_steps.append(start_idx + round(cur_idx))
62
+ cur_idx += frac_stride
63
+ all_steps += taken_steps
64
+ start_idx += size
65
+ return set(all_steps)
66
+
67
+
68
+ class SpacedDiffusion(GaussianDiffusion):
69
+ """
70
+ A diffusion process which can skip steps in a base diffusion process.
71
+
72
+ :param use_timesteps: a collection (sequence or set) of timesteps from the
73
+ original diffusion process to retain.
74
+ :param kwargs: the kwargs to create the base diffusion process.
75
+ """
76
+
77
+ def __init__(self, use_timesteps, **kwargs):
78
+ self.use_timesteps = set(use_timesteps)
79
+ self.timestep_map = []
80
+ self.original_num_steps = len(kwargs["betas"])
81
+
82
+ base_diffusion = GaussianDiffusion(**kwargs) # pylint: disable=missing-kwoa
83
+ last_alpha_cumprod = 1.0
84
+ new_betas = []
85
+ for i, alpha_cumprod in enumerate(base_diffusion.alphas_cumprod):
86
+ if i in self.use_timesteps:
87
+ new_betas.append(1 - alpha_cumprod / last_alpha_cumprod)
88
+ last_alpha_cumprod = alpha_cumprod
89
+ self.timestep_map.append(i)
90
+ kwargs["betas"] = np.array(new_betas)
91
+ super().__init__(**kwargs)
92
+
93
+ def p_mean_variance(self, model, *args, **kwargs):
94
+ return super().p_mean_variance(self._wrap_model(model), *args, **kwargs)
95
+
96
+ def condition_mean(self, cond_fn, *args, **kwargs):
97
+ return super().condition_mean(self._wrap_model(cond_fn), *args, **kwargs)
98
+
99
+ def condition_score(self, cond_fn, *args, **kwargs):
100
+ return super().condition_score(self._wrap_model(cond_fn), *args, **kwargs)
101
+
102
+ def _wrap_model(self, model):
103
+ if isinstance(model, _WrappedModel):
104
+ return model
105
+ return _WrappedModel(model, self.timestep_map, self.original_num_steps)
106
+
107
+
108
+ class _WrappedModel:
109
+ def __init__(self, model, timestep_map, original_num_steps):
110
+ self.model = model
111
+ self.timestep_map = timestep_map
112
+ self.original_num_steps = original_num_steps
113
+
114
+ def __call__(self, x, ts, **kwargs):
115
+ map_tensor = th.tensor(self.timestep_map, device=ts.device, dtype=ts.dtype)
116
+ new_ts = map_tensor[ts]
117
+ return self.model(x, new_ts, **kwargs)
glide_text2im/text2im_model.py ADDED
@@ -0,0 +1,233 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch as th
2
+ import torch.nn as nn
3
+ import torch.nn.functional as F
4
+
5
+ from .nn import timestep_embedding
6
+ from .unet import UNetModel
7
+ from .xf import LayerNorm, Transformer, convert_module_to_f16
8
+
9
+
10
+ class Text2ImUNet(UNetModel):
11
+ """
12
+ A UNetModel that conditions on text with an encoding transformer.
13
+
14
+ Expects an extra kwarg `tokens` of text.
15
+
16
+ :param text_ctx: number of text tokens to expect.
17
+ :param xf_width: width of the transformer.
18
+ :param xf_layers: depth of the transformer.
19
+ :param xf_heads: heads in the transformer.
20
+ :param xf_final_ln: use a LayerNorm after the output layer.
21
+ :param tokenizer: the text tokenizer for sampling/vocab size.
22
+ """
23
+
24
+ def __init__(
25
+ self,
26
+ text_ctx,
27
+ xf_width,
28
+ xf_layers,
29
+ xf_heads,
30
+ xf_final_ln,
31
+ tokenizer,
32
+ *args,
33
+ cache_text_emb=False,
34
+ xf_ar=0.0,
35
+ xf_padding=False,
36
+ share_unemb=False,
37
+ **kwargs,
38
+ ):
39
+ self.text_ctx = text_ctx
40
+ self.xf_width = xf_width
41
+ self.xf_ar = xf_ar
42
+ self.xf_padding = xf_padding
43
+ self.tokenizer = tokenizer
44
+
45
+ if not xf_width:
46
+ super().__init__(*args, **kwargs, encoder_channels=None)
47
+ else:
48
+ super().__init__(*args, **kwargs, encoder_channels=xf_width)
49
+ if self.xf_width:
50
+ self.transformer = Transformer(
51
+ text_ctx,
52
+ xf_width,
53
+ xf_layers,
54
+ xf_heads,
55
+ )
56
+ if xf_final_ln:
57
+ self.final_ln = LayerNorm(xf_width)
58
+ else:
59
+ self.final_ln = None
60
+
61
+ self.token_embedding = nn.Embedding(self.tokenizer.n_vocab, xf_width)
62
+ self.positional_embedding = nn.Parameter(th.empty(text_ctx, xf_width, dtype=th.float32))
63
+ self.transformer_proj = nn.Linear(xf_width, self.model_channels * 4)
64
+
65
+ if self.xf_padding:
66
+ self.padding_embedding = nn.Parameter(
67
+ th.empty(text_ctx, xf_width, dtype=th.float32)
68
+ )
69
+ if self.xf_ar:
70
+ self.unemb = nn.Linear(xf_width, self.tokenizer.n_vocab)
71
+ if share_unemb:
72
+ self.unemb.weight = self.token_embedding.weight
73
+
74
+ self.cache_text_emb = cache_text_emb
75
+ self.cache = None
76
+
77
+ def convert_to_fp16(self):
78
+ super().convert_to_fp16()
79
+ if self.xf_width:
80
+ self.transformer.apply(convert_module_to_f16)
81
+ self.transformer_proj.to(th.float16)
82
+ self.token_embedding.to(th.float16)
83
+ self.positional_embedding.to(th.float16)
84
+ if self.xf_padding:
85
+ self.padding_embedding.to(th.float16)
86
+ if self.xf_ar:
87
+ self.unemb.to(th.float16)
88
+
89
+ def get_text_emb(self, tokens, mask):
90
+ assert tokens is not None
91
+
92
+ if self.cache_text_emb and self.cache is not None:
93
+ assert (
94
+ tokens == self.cache["tokens"]
95
+ ).all(), f"Tokens {tokens.cpu().numpy().tolist()} do not match cache {self.cache['tokens'].cpu().numpy().tolist()}"
96
+ return self.cache
97
+
98
+ xf_in = self.token_embedding(tokens.long())
99
+ xf_in = xf_in + self.positional_embedding[None]
100
+ if self.xf_padding:
101
+ assert mask is not None
102
+ xf_in = th.where(mask[..., None], xf_in, self.padding_embedding[None])
103
+ xf_out = self.transformer(xf_in.to(self.dtype))
104
+ if self.final_ln is not None:
105
+ xf_out = self.final_ln(xf_out)
106
+ xf_proj = self.transformer_proj(xf_out[:, -1])
107
+ xf_out = xf_out.permute(0, 2, 1) # NLC -> NCL
108
+
109
+ outputs = dict(xf_proj=xf_proj, xf_out=xf_out)
110
+
111
+ if self.cache_text_emb:
112
+ self.cache = dict(
113
+ tokens=tokens,
114
+ xf_proj=xf_proj.detach(),
115
+ xf_out=xf_out.detach() if xf_out is not None else None,
116
+ )
117
+
118
+ return outputs
119
+
120
+ def del_cache(self):
121
+ self.cache = None
122
+
123
+ def forward(self, x, timesteps, tokens=None, mask=None):
124
+ hs = []
125
+ emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
126
+ if self.xf_width:
127
+ text_outputs = self.get_text_emb(tokens, mask)
128
+ xf_proj, xf_out = text_outputs["xf_proj"], text_outputs["xf_out"]
129
+ emb = emb + xf_proj.to(emb)
130
+ else:
131
+ xf_out = None
132
+ h = x.type(self.dtype)
133
+ for module in self.input_blocks:
134
+ h = module(h, emb, xf_out)
135
+ hs.append(h)
136
+ h = self.middle_block(h, emb, xf_out)
137
+ for module in self.output_blocks:
138
+ h = th.cat([h, hs.pop()], dim=1)
139
+ h = module(h, emb, xf_out)
140
+ h = h.type(x.dtype)
141
+ h = self.out(h)
142
+ return h
143
+
144
+
145
+ class SuperResText2ImUNet(Text2ImUNet):
146
+ """
147
+ A text2im model that performs super-resolution.
148
+ Expects an extra kwarg `low_res` to condition on a low-resolution image.
149
+ """
150
+
151
+ def __init__(self, *args, **kwargs):
152
+ if "in_channels" in kwargs:
153
+ kwargs = dict(kwargs)
154
+ kwargs["in_channels"] = kwargs["in_channels"] * 2
155
+ else:
156
+ # Curse you, Python. Or really, just curse positional arguments :|.
157
+ args = list(args)
158
+ args[1] = args[1] * 2
159
+ super().__init__(*args, **kwargs)
160
+
161
+ def forward(self, x, timesteps, low_res=None, **kwargs):
162
+ _, _, new_height, new_width = x.shape
163
+ upsampled = F.interpolate(
164
+ low_res, (new_height, new_width), mode="bilinear", align_corners=False
165
+ )
166
+ x = th.cat([x, upsampled], dim=1)
167
+ return super().forward(x, timesteps, **kwargs)
168
+
169
+
170
+ class InpaintText2ImUNet(Text2ImUNet):
171
+ """
172
+ A text2im model which can perform inpainting.
173
+ """
174
+
175
+ def __init__(self, *args, **kwargs):
176
+ if "in_channels" in kwargs:
177
+ kwargs = dict(kwargs)
178
+ kwargs["in_channels"] = kwargs["in_channels"] * 2 + 1
179
+ else:
180
+ # Curse you, Python. Or really, just curse positional arguments :|.
181
+ args = list(args)
182
+ args[1] = args[1] * 2 + 1
183
+ super().__init__(*args, **kwargs)
184
+
185
+ def forward(self, x, timesteps, inpaint_image=None, inpaint_mask=None, **kwargs):
186
+ if inpaint_image is None:
187
+ inpaint_image = th.zeros_like(x)
188
+ if inpaint_mask is None:
189
+ inpaint_mask = th.zeros_like(x[:, :1])
190
+ return super().forward(
191
+ th.cat([x, inpaint_image * inpaint_mask, inpaint_mask], dim=1),
192
+ timesteps,
193
+ **kwargs,
194
+ )
195
+
196
+
197
+ class SuperResInpaintText2ImUnet(Text2ImUNet):
198
+ """
199
+ A text2im model which can perform both upsampling and inpainting.
200
+ """
201
+
202
+ def __init__(self, *args, **kwargs):
203
+ if "in_channels" in kwargs:
204
+ kwargs = dict(kwargs)
205
+ kwargs["in_channels"] = kwargs["in_channels"] * 3 + 1
206
+ else:
207
+ # Curse you, Python. Or really, just curse positional arguments :|.
208
+ args = list(args)
209
+ args[1] = args[1] * 3 + 1
210
+ super().__init__(*args, **kwargs)
211
+
212
+ def forward(
213
+ self,
214
+ x,
215
+ timesteps,
216
+ inpaint_image=None,
217
+ inpaint_mask=None,
218
+ low_res=None,
219
+ **kwargs,
220
+ ):
221
+ if inpaint_image is None:
222
+ inpaint_image = th.zeros_like(x)
223
+ if inpaint_mask is None:
224
+ inpaint_mask = th.zeros_like(x[:, :1])
225
+ _, _, new_height, new_width = x.shape
226
+ upsampled = F.interpolate(
227
+ low_res, (new_height, new_width), mode="bilinear", align_corners=False
228
+ )
229
+ return super().forward(
230
+ th.cat([x, inpaint_image * inpaint_mask, inpaint_mask, upsampled], dim=1),
231
+ timesteps,
232
+ **kwargs,
233
+ )
glide_text2im/tokenizer/__init__.py ADDED
File without changes
glide_text2im/tokenizer/bpe.py ADDED
@@ -0,0 +1,151 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Byte pair encoding utilities adapted from:
3
+ https://github.com/openai/gpt-2/blob/master/src/encoder.py
4
+ """
5
+
6
+ import gzip
7
+ import json
8
+ import os
9
+ from functools import lru_cache
10
+ from typing import List, Tuple
11
+
12
+ import regex as re
13
+
14
+
15
+ @lru_cache()
16
+ def bytes_to_unicode():
17
+ """
18
+ Returns list of utf-8 byte and a corresponding list of unicode strings.
19
+ The reversible bpe codes work on unicode strings.
20
+ This means you need a large # of unicode characters in your vocab if you want to avoid UNKs.
21
+ When you're at something like a 10B token dataset you end up needing around 5K for decent coverage.
22
+ This is a signficant percentage of your normal, say, 32K bpe vocab.
23
+ To avoid that, we want lookup tables between utf-8 bytes and unicode strings.
24
+ And avoids mapping to whitespace/control characters the bpe code barfs on.
25
+ """
26
+ bs = (
27
+ list(range(ord("!"), ord("~") + 1))
28
+ + list(range(ord("¡"), ord("¬") + 1))
29
+ + list(range(ord("®"), ord("ÿ") + 1))
30
+ )
31
+ cs = bs[:]
32
+ n = 0
33
+ for b in range(2 ** 8):
34
+ if b not in bs:
35
+ bs.append(b)
36
+ cs.append(2 ** 8 + n)
37
+ n += 1
38
+ cs = [chr(n) for n in cs]
39
+ return dict(zip(bs, cs))
40
+
41
+
42
+ def get_pairs(word):
43
+ """Return set of symbol pairs in a word.
44
+ Word is represented as tuple of symbols (symbols being variable-length strings).
45
+ """
46
+ pairs = set()
47
+ prev_char = word[0]
48
+ for char in word[1:]:
49
+ pairs.add((prev_char, char))
50
+ prev_char = char
51
+ return pairs
52
+
53
+
54
+ class Encoder:
55
+ def __init__(self, encoder, bpe_merges, errors="replace"):
56
+ self.encoder = encoder
57
+ self.decoder = {v: k for k, v in self.encoder.items()}
58
+ self.errors = errors # how to handle errors in decoding
59
+ self.byte_encoder = bytes_to_unicode()
60
+ self.byte_decoder = {v: k for k, v in self.byte_encoder.items()}
61
+ self.bpe_ranks = dict(zip(bpe_merges, range(len(bpe_merges))))
62
+ self.cache = {}
63
+
64
+ # Should haved added re.IGNORECASE so BPE merges can happen for capitalized versions of contractions
65
+ self.pat = re.compile(
66
+ r"""'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+"""
67
+ )
68
+
69
+ @property
70
+ def n_vocab(self) -> int:
71
+ return len(self.encoder)
72
+
73
+ @property
74
+ def end_token(self) -> int:
75
+ return self.n_vocab - 1
76
+
77
+ def padded_tokens_and_mask(
78
+ self, tokens: List[int], text_ctx: int
79
+ ) -> Tuple[List[int], List[bool]]:
80
+ tokens = tokens[:text_ctx]
81
+ padding = text_ctx - len(tokens)
82
+ padded_tokens = tokens + [self.end_token] * padding
83
+ mask = [True] * len(tokens) + [False] * padding
84
+ return padded_tokens, mask
85
+
86
+ def bpe(self, token):
87
+ if token in self.cache:
88
+ return self.cache[token]
89
+ word = tuple(token)
90
+ pairs = get_pairs(word)
91
+
92
+ if not pairs:
93
+ return token
94
+
95
+ while True:
96
+ bigram = min(pairs, key=lambda pair: self.bpe_ranks.get(pair, float("inf")))
97
+ if bigram not in self.bpe_ranks:
98
+ break
99
+ first, second = bigram
100
+ new_word = []
101
+ i = 0
102
+ while i < len(word):
103
+ try:
104
+ j = word.index(first, i)
105
+ new_word.extend(word[i:j])
106
+ i = j
107
+ except: # pylint: disable=bare-except
108
+ new_word.extend(word[i:])
109
+ break
110
+
111
+ if word[i] == first and i < len(word) - 1 and word[i + 1] == second:
112
+ new_word.append(first + second)
113
+ i += 2
114
+ else:
115
+ new_word.append(word[i])
116
+ i += 1
117
+ new_word = tuple(new_word)
118
+ word = new_word
119
+ if len(word) == 1:
120
+ break
121
+ else:
122
+ pairs = get_pairs(word)
123
+ word = " ".join(word)
124
+ self.cache[token] = word
125
+ return word
126
+
127
+ def encode(self, text):
128
+ text = text.lower()
129
+ bpe_tokens = []
130
+ for token in re.findall(self.pat, text):
131
+ token = "".join(self.byte_encoder[b] for b in token.encode("utf-8"))
132
+ bpe_tokens.extend(self.encoder[bpe_token] for bpe_token in self.bpe(token).split(" "))
133
+ return bpe_tokens
134
+
135
+ def decode(self, tokens):
136
+ text = "".join([self.decoder[token] for token in tokens])
137
+ text = bytearray([self.byte_decoder[c] for c in text]).decode("utf-8", errors=self.errors)
138
+ return text
139
+
140
+
141
+ def get_encoder():
142
+ root_dir = os.path.dirname(os.path.abspath(__file__))
143
+ with gzip.open(os.path.join(root_dir, "encoder.json.gz"), "r") as f:
144
+ encoder = json.load(f)
145
+ with gzip.open(os.path.join(root_dir, "vocab.bpe.gz"), "r") as f:
146
+ bpe_data = str(f.read(), "utf-8")
147
+ bpe_merges = [tuple(merge_str.split()) for merge_str in bpe_data.split("\n")[1:-1]]
148
+ return Encoder(
149
+ encoder=encoder,
150
+ bpe_merges=bpe_merges,
151
+ )
glide_text2im/tokenizer/bpe_simple_vocab_16e6.txt.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:924691ac288e54409236115652ad4aa250f48203de50a9e4722a6ecd48d6804a
3
+ size 1356917
glide_text2im/tokenizer/encoder.json.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4debc1cf25180021b07744bc9f4488d53c7bf112c8ce5de8097c6a7518f4ec7c
3
+ size 348346
glide_text2im/tokenizer/simple_tokenizer.py ADDED
@@ -0,0 +1,163 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Copied from: https://github.com/openai/CLIP/blob/573315e83f07b53a61ff5098757e8fc885f1703e/clip/simple_tokenizer.py
3
+ """
4
+
5
+ import gzip
6
+ import html
7
+ import os
8
+ from functools import lru_cache
9
+ from typing import List, Tuple
10
+
11
+ import ftfy
12
+ import regex as re
13
+
14
+
15
+ @lru_cache()
16
+ def default_bpe():
17
+ return os.path.join(os.path.dirname(os.path.abspath(__file__)), "bpe_simple_vocab_16e6.txt.gz")
18
+
19
+
20
+ @lru_cache()
21
+ def bytes_to_unicode():
22
+ """
23
+ Returns list of utf-8 byte and a corresponding list of unicode strings.
24
+ The reversible bpe codes work on unicode strings.
25
+ This means you need a large # of unicode characters in your vocab if you want to avoid UNKs.
26
+ When you're at something like a 10B token dataset you end up needing around 5K for decent coverage.
27
+ This is a signficant percentage of your normal, say, 32K bpe vocab.
28
+ To avoid that, we want lookup tables between utf-8 bytes and unicode strings.
29
+ And avoids mapping to whitespace/control characters the bpe code barfs on.
30
+ """
31
+ bs = (
32
+ list(range(ord("!"), ord("~") + 1))
33
+ + list(range(ord("¡"), ord("¬") + 1))
34
+ + list(range(ord("®"), ord("ÿ") + 1))
35
+ )
36
+ cs = bs[:]
37
+ n = 0
38
+ for b in range(2 ** 8):
39
+ if b not in bs:
40
+ bs.append(b)
41
+ cs.append(2 ** 8 + n)
42
+ n += 1
43
+ cs = [chr(n) for n in cs]
44
+ return dict(zip(bs, cs))
45
+
46
+
47
+ def get_pairs(word):
48
+ """Return set of symbol pairs in a word.
49
+ Word is represented as tuple of symbols (symbols being variable-length strings).
50
+ """
51
+ pairs = set()
52
+ prev_char = word[0]
53
+ for char in word[1:]:
54
+ pairs.add((prev_char, char))
55
+ prev_char = char
56
+ return pairs
57
+
58
+
59
+ def basic_clean(text):
60
+ text = ftfy.fix_text(text)
61
+ text = html.unescape(html.unescape(text))
62
+ return text.strip()
63
+
64
+
65
+ def whitespace_clean(text):
66
+ text = re.sub(r"\s+", " ", text)
67
+ text = text.strip()
68
+ return text
69
+
70
+
71
+ class SimpleTokenizer(object):
72
+ def __init__(self, bpe_path: str = default_bpe()):
73
+ self.byte_encoder = bytes_to_unicode()
74
+ self.byte_decoder = {v: k for k, v in self.byte_encoder.items()}
75
+ merges = gzip.open(bpe_path).read().decode("utf-8").split("\n")
76
+ merges = merges[1 : 49152 - 256 - 2 + 1]
77
+ merges = [tuple(merge.split()) for merge in merges]
78
+ vocab = list(bytes_to_unicode().values())
79
+ vocab = vocab + [v + "</w>" for v in vocab]
80
+ for merge in merges:
81
+ vocab.append("".join(merge))
82
+ vocab.extend(["<|startoftext|>", "<|endoftext|>"])
83
+ self.encoder = dict(zip(vocab, range(len(vocab))))
84
+ self.decoder = {v: k for k, v in self.encoder.items()}
85
+ self.bpe_ranks = dict(zip(merges, range(len(merges))))
86
+ self.cache = {"<|startoftext|>": "<|startoftext|>", "<|endoftext|>": "<|endoftext|>"}
87
+ self.pat = re.compile(
88
+ r"""<\|startoftext\|>|<\|endoftext\|>|'s|'t|'re|'ve|'m|'ll|'d|[\p{L}]+|[\p{N}]|[^\s\p{L}\p{N}]+""",
89
+ re.IGNORECASE,
90
+ )
91
+
92
+ @property
93
+ def start_token(self):
94
+ return self.encoder["<|startoftext|>"]
95
+
96
+ @property
97
+ def end_token(self):
98
+ return self.encoder["<|endoftext|>"]
99
+
100
+ def padded_tokens_and_len(self, tokens: List[int], text_ctx: int) -> Tuple[List[int], int]:
101
+ tokens = [self.start_token] + tokens[: text_ctx - 2] + [self.end_token]
102
+ text_len = len(tokens)
103
+ padding = text_ctx - len(tokens)
104
+ padded_tokens = tokens + [0] * padding
105
+ return padded_tokens, text_len
106
+
107
+ def bpe(self, token):
108
+ if token in self.cache:
109
+ return self.cache[token]
110
+ word = tuple(token[:-1]) + (token[-1] + "</w>",)
111
+ pairs = get_pairs(word)
112
+
113
+ if not pairs:
114
+ return token + "</w>"
115
+
116
+ while True:
117
+ bigram = min(pairs, key=lambda pair: self.bpe_ranks.get(pair, float("inf")))
118
+ if bigram not in self.bpe_ranks:
119
+ break
120
+ first, second = bigram
121
+ new_word = []
122
+ i = 0
123
+ while i < len(word):
124
+ try:
125
+ j = word.index(first, i)
126
+ new_word.extend(word[i:j])
127
+ i = j
128
+ except: # pylint: disable=bare-except
129
+ new_word.extend(word[i:])
130
+ break
131
+
132
+ if word[i] == first and i < len(word) - 1 and word[i + 1] == second:
133
+ new_word.append(first + second)
134
+ i += 2
135
+ else:
136
+ new_word.append(word[i])
137
+ i += 1
138
+ new_word = tuple(new_word)
139
+ word = new_word
140
+ if len(word) == 1:
141
+ break
142
+ else:
143
+ pairs = get_pairs(word)
144
+ word = " ".join(word)
145
+ self.cache[token] = word
146
+ return word
147
+
148
+ def encode(self, text):
149
+ bpe_tokens = []
150
+ text = whitespace_clean(basic_clean(text)).lower()
151
+ for token in re.findall(self.pat, text):
152
+ token = "".join(self.byte_encoder[b] for b in token.encode("utf-8"))
153
+ bpe_tokens.extend(self.encoder[bpe_token] for bpe_token in self.bpe(token).split(" "))
154
+ return bpe_tokens
155
+
156
+ def decode(self, tokens):
157
+ text = "".join([self.decoder[token] for token in tokens])
158
+ text = (
159
+ bytearray([self.byte_decoder[c] for c in text])
160
+ .decode("utf-8", errors="replace")
161
+ .replace("</w>", " ")
162
+ )
163
+ return text
glide_text2im/tokenizer/vocab.bpe.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce239dd5a898827423fee00e3f7ab37de7900f247f2ba360753d860e8a46524d
3
+ size 213544
glide_text2im/unet.py ADDED
@@ -0,0 +1,635 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import math
2
+ from abc import abstractmethod
3
+
4
+ import torch as th
5
+ import torch.nn as nn
6
+ import torch.nn.functional as F
7
+
8
+ from .fp16_util import convert_module_to_f16, convert_module_to_f32
9
+ from .nn import avg_pool_nd, conv_nd, linear, normalization, timestep_embedding, zero_module
10
+
11
+
12
+ class TimestepBlock(nn.Module):
13
+ """
14
+ Any module where forward() takes timestep embeddings as a second argument.
15
+ """
16
+
17
+ @abstractmethod
18
+ def forward(self, x, emb):
19
+ """
20
+ Apply the module to `x` given `emb` timestep embeddings.
21
+ """
22
+
23
+
24
+ class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
25
+ """
26
+ A sequential module that passes timestep embeddings to the children that
27
+ support it as an extra input.
28
+ """
29
+
30
+ def forward(self, x, emb, encoder_out=None):
31
+ for layer in self:
32
+ if isinstance(layer, TimestepBlock):
33
+ x = layer(x, emb)
34
+ elif isinstance(layer, AttentionBlock):
35
+ x = layer(x, encoder_out)
36
+ else:
37
+ x = layer(x)
38
+ return x
39
+
40
+
41
+ class Upsample(nn.Module):
42
+ """
43
+ An upsampling layer with an optional convolution.
44
+
45
+ :param channels: channels in the inputs and outputs.
46
+ :param use_conv: a bool determining if a convolution is applied.
47
+ :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
48
+ upsampling occurs in the inner-two dimensions.
49
+ """
50
+
51
+ def __init__(self, channels, use_conv, dims=2, out_channels=None):
52
+ super().__init__()
53
+ self.channels = channels
54
+ self.out_channels = out_channels or channels
55
+ self.use_conv = use_conv
56
+ self.dims = dims
57
+ if use_conv:
58
+ self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=1)
59
+
60
+ def forward(self, x):
61
+ assert x.shape[1] == self.channels
62
+ if self.dims == 3:
63
+ x = F.interpolate(x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest")
64
+ else:
65
+ x = F.interpolate(x, scale_factor=2, mode="nearest")
66
+ if self.use_conv:
67
+ x = self.conv(x)
68
+ return x
69
+
70
+
71
+ class Downsample(nn.Module):
72
+ """
73
+ A downsampling layer with an optional convolution.
74
+
75
+ :param channels: channels in the inputs and outputs.
76
+ :param use_conv: a bool determining if a convolution is applied.
77
+ :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
78
+ downsampling occurs in the inner-two dimensions.
79
+ """
80
+
81
+ def __init__(self, channels, use_conv, dims=2, out_channels=None):
82
+ super().__init__()
83
+ self.channels = channels
84
+ self.out_channels = out_channels or channels
85
+ self.use_conv = use_conv
86
+ self.dims = dims
87
+ stride = 2 if dims != 3 else (1, 2, 2)
88
+ if use_conv:
89
+ self.op = conv_nd(dims, self.channels, self.out_channels, 3, stride=stride, padding=1)
90
+ else:
91
+ assert self.channels == self.out_channels
92
+ self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
93
+
94
+ def forward(self, x):
95
+ assert x.shape[1] == self.channels
96
+ return self.op(x)
97
+
98
+
99
+ class ResBlock(TimestepBlock):
100
+ """
101
+ A residual block that can optionally change the number of channels.
102
+
103
+ :param channels: the number of input channels.
104
+ :param emb_channels: the number of timestep embedding channels.
105
+ :param dropout: the rate of dropout.
106
+ :param out_channels: if specified, the number of out channels.
107
+ :param use_conv: if True and out_channels is specified, use a spatial
108
+ convolution instead of a smaller 1x1 convolution to change the
109
+ channels in the skip connection.
110
+ :param dims: determines if the signal is 1D, 2D, or 3D.
111
+ :param use_checkpoint: if True, use gradient checkpointing on this module.
112
+ :param up: if True, use this block for upsampling.
113
+ :param down: if True, use this block for downsampling.
114
+ """
115
+
116
+ def __init__(
117
+ self,
118
+ channels,
119
+ emb_channels,
120
+ dropout,
121
+ out_channels=None,
122
+ use_conv=False,
123
+ use_scale_shift_norm=False,
124
+ dims=2,
125
+ use_checkpoint=False,
126
+ up=False,
127
+ down=False,
128
+ ):
129
+ super().__init__()
130
+ self.channels = channels
131
+ self.emb_channels = emb_channels
132
+ self.dropout = dropout
133
+ self.out_channels = out_channels or channels
134
+ self.use_conv = use_conv
135
+ self.use_checkpoint = use_checkpoint
136
+ self.use_scale_shift_norm = use_scale_shift_norm
137
+
138
+ self.in_layers = nn.Sequential(
139
+ normalization(channels, swish=1.0),
140
+ nn.Identity(),
141
+ conv_nd(dims, channels, self.out_channels, 3, padding=1),
142
+ )
143
+
144
+ self.updown = up or down
145
+
146
+ if up:
147
+ self.h_upd = Upsample(channels, False, dims)
148
+ self.x_upd = Upsample(channels, False, dims)
149
+ elif down:
150
+ self.h_upd = Downsample(channels, False, dims)
151
+ self.x_upd = Downsample(channels, False, dims)
152
+ else:
153
+ self.h_upd = self.x_upd = nn.Identity()
154
+
155
+ self.emb_layers = nn.Sequential(
156
+ nn.SiLU(),
157
+ linear(
158
+ emb_channels,
159
+ 2 * self.out_channels if use_scale_shift_norm else self.out_channels,
160
+ ),
161
+ )
162
+ self.out_layers = nn.Sequential(
163
+ normalization(self.out_channels, swish=0.0 if use_scale_shift_norm else 1.0),
164
+ nn.SiLU() if use_scale_shift_norm else nn.Identity(),
165
+ nn.Dropout(p=dropout),
166
+ zero_module(conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)),
167
+ )
168
+
169
+ if self.out_channels == channels:
170
+ self.skip_connection = nn.Identity()
171
+ elif use_conv:
172
+ self.skip_connection = conv_nd(dims, channels, self.out_channels, 3, padding=1)
173
+ else:
174
+ self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
175
+
176
+ def forward(self, x, emb):
177
+ """
178
+ Apply the block to a Tensor, conditioned on a timestep embedding.
179
+
180
+ :param x: an [N x C x ...] Tensor of features.
181
+ :param emb: an [N x emb_channels] Tensor of timestep embeddings.
182
+ :return: an [N x C x ...] Tensor of outputs.
183
+ """
184
+ if self.updown:
185
+ in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
186
+ h = in_rest(x)
187
+ h = self.h_upd(h)
188
+ x = self.x_upd(x)
189
+ h = in_conv(h)
190
+ else:
191
+ h = self.in_layers(x)
192
+ emb_out = self.emb_layers(emb).type(h.dtype)
193
+ while len(emb_out.shape) < len(h.shape):
194
+ emb_out = emb_out[..., None]
195
+ if self.use_scale_shift_norm:
196
+ out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
197
+ scale, shift = th.chunk(emb_out, 2, dim=1)
198
+ h = out_norm(h) * (1 + scale) + shift
199
+ h = out_rest(h)
200
+ else:
201
+ h = h + emb_out
202
+ h = self.out_layers(h)
203
+ return self.skip_connection(x) + h
204
+
205
+
206
+ class AttentionBlock(nn.Module):
207
+ """
208
+ An attention block that allows spatial positions to attend to each other.
209
+
210
+ Originally ported from here, but adapted to the N-d case.
211
+ https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66.
212
+ """
213
+
214
+ def __init__(
215
+ self,
216
+ channels,
217
+ num_heads=1,
218
+ num_head_channels=-1,
219
+ use_checkpoint=False,
220
+ encoder_channels=None,
221
+ ):
222
+ super().__init__()
223
+ self.channels = channels
224
+ if num_head_channels == -1:
225
+ self.num_heads = num_heads
226
+ else:
227
+ assert (
228
+ channels % num_head_channels == 0
229
+ ), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
230
+ self.num_heads = channels // num_head_channels
231
+ self.use_checkpoint = use_checkpoint
232
+ self.norm = normalization(channels, swish=0.0)
233
+ self.qkv = conv_nd(1, channels, channels * 3, 1)
234
+ self.attention = QKVAttention(self.num_heads)
235
+
236
+ if encoder_channels is not None:
237
+ self.encoder_kv = conv_nd(1, encoder_channels, channels * 2, 1)
238
+ self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
239
+
240
+ def forward(self, x, encoder_out=None):
241
+ b, c, *spatial = x.shape
242
+ qkv = self.qkv(self.norm(x).view(b, c, -1))
243
+ if encoder_out is not None:
244
+ encoder_out = self.encoder_kv(encoder_out)
245
+ h = self.attention(qkv, encoder_out)
246
+ else:
247
+ h = self.attention(qkv)
248
+ h = self.proj_out(h)
249
+ return x + h.reshape(b, c, *spatial)
250
+
251
+
252
+ class QKVAttention(nn.Module):
253
+ """
254
+ A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
255
+ """
256
+
257
+ def __init__(self, n_heads):
258
+ super().__init__()
259
+ self.n_heads = n_heads
260
+
261
+ def forward(self, qkv, encoder_kv=None):
262
+ """
263
+ Apply QKV attention.
264
+
265
+ :param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
266
+ :return: an [N x (H * C) x T] tensor after attention.
267
+ """
268
+ bs, width, length = qkv.shape
269
+ assert width % (3 * self.n_heads) == 0
270
+ ch = width // (3 * self.n_heads)
271
+ q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
272
+ if encoder_kv is not None:
273
+ assert encoder_kv.shape[1] == self.n_heads * ch * 2
274
+ ek, ev = encoder_kv.reshape(bs * self.n_heads, ch * 2, -1).split(ch, dim=1)
275
+ k = th.cat([ek, k], dim=-1)
276
+ v = th.cat([ev, v], dim=-1)
277
+ scale = 1 / math.sqrt(math.sqrt(ch))
278
+ weight = th.einsum(
279
+ "bct,bcs->bts", q * scale, k * scale
280
+ ) # More stable with f16 than dividing afterwards
281
+ weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
282
+ a = th.einsum("bts,bcs->bct", weight, v)
283
+ return a.reshape(bs, -1, length)
284
+
285
+
286
+ class UNetModel(nn.Module):
287
+ """
288
+ The full UNet model with attention and timestep embedding.
289
+
290
+ :param in_channels: channels in the input Tensor.
291
+ :param model_channels: base channel count for the model.
292
+ :param out_channels: channels in the output Tensor.
293
+ :param num_res_blocks: number of residual blocks per downsample.
294
+ :param attention_resolutions: a collection of downsample rates at which
295
+ attention will take place. May be a set, list, or tuple.
296
+ For example, if this contains 4, then at 4x downsampling, attention
297
+ will be used.
298
+ :param dropout: the dropout probability.
299
+ :param channel_mult: channel multiplier for each level of the UNet.
300
+ :param conv_resample: if True, use learned convolutions for upsampling and
301
+ downsampling.
302
+ :param dims: determines if the signal is 1D, 2D, or 3D.
303
+ :param num_classes: if specified (as an int), then this model will be
304
+ class-conditional with `num_classes` classes.
305
+ :param use_checkpoint: use gradient checkpointing to reduce memory usage.
306
+ :param num_heads: the number of attention heads in each attention layer.
307
+ :param num_heads_channels: if specified, ignore num_heads and instead use
308
+ a fixed channel width per attention head.
309
+ :param num_heads_upsample: works with num_heads to set a different number
310
+ of heads for upsampling. Deprecated.
311
+ :param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
312
+ :param resblock_updown: use residual blocks for up/downsampling.
313
+ """
314
+
315
+ def __init__(
316
+ self,
317
+ in_channels,
318
+ model_channels,
319
+ out_channels,
320
+ num_res_blocks,
321
+ attention_resolutions,
322
+ dropout=0,
323
+ channel_mult=(1, 2, 4, 8),
324
+ conv_resample=True,
325
+ dims=2,
326
+ num_classes=None,
327
+ use_checkpoint=False,
328
+ use_fp16=False,
329
+ num_heads=1,
330
+ num_head_channels=-1,
331
+ num_heads_upsample=-1,
332
+ use_scale_shift_norm=False,
333
+ resblock_updown=False,
334
+ encoder_channels=None,
335
+ ):
336
+ super().__init__()
337
+
338
+ if num_heads_upsample == -1:
339
+ num_heads_upsample = num_heads
340
+
341
+ self.in_channels = in_channels
342
+ self.model_channels = model_channels
343
+ self.out_channels = out_channels
344
+ self.num_res_blocks = num_res_blocks
345
+ self.attention_resolutions = attention_resolutions
346
+ self.dropout = dropout
347
+ self.channel_mult = channel_mult
348
+ self.conv_resample = conv_resample
349
+ self.num_classes = num_classes
350
+ self.use_checkpoint = use_checkpoint
351
+ self.dtype = th.float16 if use_fp16 else th.float32
352
+ self.num_heads = num_heads
353
+ self.num_head_channels = num_head_channels
354
+ self.num_heads_upsample = num_heads_upsample
355
+
356
+ time_embed_dim = model_channels * 4
357
+ self.time_embed = nn.Sequential(
358
+ linear(model_channels, time_embed_dim),
359
+ nn.SiLU(),
360
+ linear(time_embed_dim, time_embed_dim),
361
+ )
362
+
363
+ if self.num_classes is not None:
364
+ self.label_emb = nn.Embedding(num_classes, time_embed_dim)
365
+
366
+ ch = input_ch = int(channel_mult[0] * model_channels)
367
+ self.input_blocks = nn.ModuleList(
368
+ [TimestepEmbedSequential(conv_nd(dims, in_channels, ch, 3, padding=1))]
369
+ )
370
+ self._feature_size = ch
371
+ input_block_chans = [ch]
372
+ ds = 1
373
+ for level, mult in enumerate(channel_mult):
374
+ for _ in range(num_res_blocks):
375
+ layers = [
376
+ ResBlock(
377
+ ch,
378
+ time_embed_dim,
379
+ dropout,
380
+ out_channels=int(mult * model_channels),
381
+ dims=dims,
382
+ use_checkpoint=use_checkpoint,
383
+ use_scale_shift_norm=use_scale_shift_norm,
384
+ )
385
+ ]
386
+ ch = int(mult * model_channels)
387
+ if ds in attention_resolutions:
388
+ layers.append(
389
+ AttentionBlock(
390
+ ch,
391
+ use_checkpoint=use_checkpoint,
392
+ num_heads=num_heads,
393
+ num_head_channels=num_head_channels,
394
+ encoder_channels=encoder_channels,
395
+ )
396
+ )
397
+ self.input_blocks.append(TimestepEmbedSequential(*layers))
398
+ self._feature_size += ch
399
+ input_block_chans.append(ch)
400
+ if level != len(channel_mult) - 1:
401
+ out_ch = ch
402
+ self.input_blocks.append(
403
+ TimestepEmbedSequential(
404
+ ResBlock(
405
+ ch,
406
+ time_embed_dim,
407
+ dropout,
408
+ out_channels=out_ch,
409
+ dims=dims,
410
+ use_checkpoint=use_checkpoint,
411
+ use_scale_shift_norm=use_scale_shift_norm,
412
+ down=True,
413
+ )
414
+ if resblock_updown
415
+ else Downsample(ch, conv_resample, dims=dims, out_channels=out_ch)
416
+ )
417
+ )
418
+ ch = out_ch
419
+ input_block_chans.append(ch)
420
+ ds *= 2
421
+ self._feature_size += ch
422
+
423
+ self.middle_block = TimestepEmbedSequential(
424
+ ResBlock(
425
+ ch,
426
+ time_embed_dim,
427
+ dropout,
428
+ dims=dims,
429
+ use_checkpoint=use_checkpoint,
430
+ use_scale_shift_norm=use_scale_shift_norm,
431
+ ),
432
+ AttentionBlock(
433
+ ch,
434
+ use_checkpoint=use_checkpoint,
435
+ num_heads=num_heads,
436
+ num_head_channels=num_head_channels,
437
+ encoder_channels=encoder_channels,
438
+ ),
439
+ ResBlock(
440
+ ch,
441
+ time_embed_dim,
442
+ dropout,
443
+ dims=dims,
444
+ use_checkpoint=use_checkpoint,
445
+ use_scale_shift_norm=use_scale_shift_norm,
446
+ ),
447
+ )
448
+ self._feature_size += ch
449
+
450
+ self.output_blocks = nn.ModuleList([])
451
+ for level, mult in list(enumerate(channel_mult))[::-1]:
452
+ for i in range(num_res_blocks + 1):
453
+ ich = input_block_chans.pop()
454
+ layers = [
455
+ ResBlock(
456
+ ch + ich,
457
+ time_embed_dim,
458
+ dropout,
459
+ out_channels=int(model_channels * mult),
460
+ dims=dims,
461
+ use_checkpoint=use_checkpoint,
462
+ use_scale_shift_norm=use_scale_shift_norm,
463
+ )
464
+ ]
465
+ ch = int(model_channels * mult)
466
+ if ds in attention_resolutions:
467
+ layers.append(
468
+ AttentionBlock(
469
+ ch,
470
+ use_checkpoint=use_checkpoint,
471
+ num_heads=num_heads_upsample,
472
+ num_head_channels=num_head_channels,
473
+ encoder_channels=encoder_channels,
474
+ )
475
+ )
476
+ if level and i == num_res_blocks:
477
+ out_ch = ch
478
+ layers.append(
479
+ ResBlock(
480
+ ch,
481
+ time_embed_dim,
482
+ dropout,
483
+ out_channels=out_ch,
484
+ dims=dims,
485
+ use_checkpoint=use_checkpoint,
486
+ use_scale_shift_norm=use_scale_shift_norm,
487
+ up=True,
488
+ )
489
+ if resblock_updown
490
+ else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
491
+ )
492
+ ds //= 2
493
+ self.output_blocks.append(TimestepEmbedSequential(*layers))
494
+ self._feature_size += ch
495
+
496
+ self.out = nn.Sequential(
497
+ normalization(ch, swish=1.0),
498
+ nn.Identity(),
499
+ zero_module(conv_nd(dims, input_ch, out_channels, 3, padding=1)),
500
+ )
501
+ self.use_fp16 = use_fp16
502
+
503
+ def convert_to_fp16(self):
504
+ """
505
+ Convert the torso of the model to float16.
506
+ """
507
+ self.input_blocks.apply(convert_module_to_f16)
508
+ self.middle_block.apply(convert_module_to_f16)
509
+ self.output_blocks.apply(convert_module_to_f16)
510
+
511
+ def convert_to_fp32(self):
512
+ """
513
+ Convert the torso of the model to float32.
514
+ """
515
+ self.input_blocks.apply(convert_module_to_f32)
516
+ self.middle_block.apply(convert_module_to_f32)
517
+ self.output_blocks.apply(convert_module_to_f32)
518
+
519
+ def forward(self, x, timesteps, y=None):
520
+ """
521
+ Apply the model to an input batch.
522
+
523
+ :param x: an [N x C x ...] Tensor of inputs.
524
+ :param timesteps: a 1-D batch of timesteps.
525
+ :param y: an [N] Tensor of labels, if class-conditional.
526
+ :return: an [N x C x ...] Tensor of outputs.
527
+ """
528
+ assert (y is not None) == (
529
+ self.num_classes is not None
530
+ ), "must specify y if and only if the model is class-conditional"
531
+
532
+ hs = []
533
+ emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
534
+
535
+ if self.num_classes is not None:
536
+ assert y.shape == (x.shape[0],)
537
+ emb = emb + self.label_emb(y)
538
+
539
+ h = x.type(self.dtype)
540
+ for module in self.input_blocks:
541
+ h = module(h, emb)
542
+ hs.append(h)
543
+ h = self.middle_block(h, emb)
544
+ for module in self.output_blocks:
545
+ h = th.cat([h, hs.pop()], dim=1)
546
+ h = module(h, emb)
547
+ h = h.type(x.dtype)
548
+ return self.out(h)
549
+
550
+ class SuperResUNetModel(UNetModel):
551
+ """
552
+ A UNetModel that performs super-resolution.
553
+
554
+ Expects an extra kwarg `low_res` to condition on a low-resolution image.
555
+ """
556
+
557
+ def __init__(self, *args, **kwargs):
558
+ if "in_channels" in kwargs:
559
+ kwargs = dict(kwargs)
560
+ kwargs["in_channels"] = kwargs["in_channels"] * 2
561
+ else:
562
+ # Curse you, Python. Or really, just curse positional arguments :|.
563
+ args = list(args)
564
+ args[1] = args[1] * 2
565
+ super().__init__(*args, **kwargs)
566
+
567
+ def forward(self, x, timesteps, low_res=None, **kwargs):
568
+ _, _, new_height, new_width = x.shape
569
+ upsampled = F.interpolate(low_res, (new_height, new_width), mode="bilinear")
570
+ x = th.cat([x, upsampled], dim=1)
571
+ return super().forward(x, timesteps, **kwargs)
572
+
573
+
574
+ class InpaintUNetModel(UNetModel):
575
+ """
576
+ A UNetModel which can perform inpainting.
577
+ """
578
+
579
+ def __init__(self, *args, **kwargs):
580
+ if "in_channels" in kwargs:
581
+ kwargs = dict(kwargs)
582
+ kwargs["in_channels"] = kwargs["in_channels"] * 2 + 1
583
+ else:
584
+ # Curse you, Python. Or really, just curse positional arguments :|.
585
+ args = list(args)
586
+ args[1] = args[1] * 2 + 1
587
+ super().__init__(*args, **kwargs)
588
+
589
+ def forward(self, x, timesteps, inpaint_image=None, inpaint_mask=None, **kwargs):
590
+ if inpaint_image is None:
591
+ inpaint_image = th.zeros_like(x)
592
+ if inpaint_mask is None:
593
+ inpaint_mask = th.zeros_like(x[:, :1])
594
+ return super().forward(
595
+ th.cat([x, inpaint_image * inpaint_mask, inpaint_mask], dim=1),
596
+ timesteps,
597
+ **kwargs,
598
+ )
599
+
600
+
601
+ class SuperResInpaintUNetModel(UNetModel):
602
+ """
603
+ A UNetModel which can perform both upsampling and inpainting.
604
+ """
605
+
606
+ def __init__(self, *args, **kwargs):
607
+ if "in_channels" in kwargs:
608
+ kwargs = dict(kwargs)
609
+ kwargs["in_channels"] = kwargs["in_channels"] * 3 + 1
610
+ else:
611
+ # Curse you, Python. Or really, just curse positional arguments :|.
612
+ args = list(args)
613
+ args[1] = args[1] * 3 + 1
614
+ super().__init__(*args, **kwargs)
615
+
616
+ def forward(
617
+ self,
618
+ x,
619
+ timesteps,
620
+ inpaint_image=None,
621
+ inpaint_mask=None,
622
+ low_res=None,
623
+ **kwargs,
624
+ ):
625
+ if inpaint_image is None:
626
+ inpaint_image = th.zeros_like(x)
627
+ if inpaint_mask is None:
628
+ inpaint_mask = th.zeros_like(x[:, :1])
629
+ _, _, new_height, new_width = x.shape
630
+ upsampled = F.interpolate(low_res, (new_height, new_width), mode="bilinear")
631
+ return super().forward(
632
+ th.cat([x, inpaint_image * inpaint_mask, inpaint_mask, upsampled], dim=1),
633
+ timesteps,
634
+ **kwargs,
635
+ )
glide_text2im/xf.py ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Transformer implementation adapted from CLIP ViT:
3
+ https://github.com/openai/CLIP/blob/4c0275784d6d9da97ca1f47eaaee31de1867da91/clip/model.py
4
+ """
5
+
6
+ import math
7
+
8
+ import torch as th
9
+ import torch.nn as nn
10
+
11
+
12
+ def convert_module_to_f16(l):
13
+ """
14
+ Convert primitive modules to float16.
15
+ """
16
+ if isinstance(l, (nn.Linear, nn.Conv2d, nn.ConvTranspose2d)):
17
+ l.weight.data = l.weight.data.half()
18
+ if l.bias is not None:
19
+ l.bias.data = l.bias.data.half()
20
+
21
+
22
+ class LayerNorm(nn.LayerNorm):
23
+ """
24
+ Implementation that supports fp16 inputs but fp32 gains/biases.
25
+ """
26
+
27
+ def forward(self, x: th.Tensor):
28
+ return super().forward(x.float()).to(x.dtype)
29
+
30
+
31
+ class MultiheadAttention(nn.Module):
32
+ def __init__(self, n_ctx, width, heads):
33
+ super().__init__()
34
+ self.n_ctx = n_ctx
35
+ self.width = width
36
+ self.heads = heads
37
+ self.c_qkv = nn.Linear(width, width * 3)
38
+ self.c_proj = nn.Linear(width, width)
39
+ self.attention = QKVMultiheadAttention(heads, n_ctx)
40
+
41
+ def forward(self, x):
42
+ x = self.c_qkv(x)
43
+ x = self.attention(x)
44
+ x = self.c_proj(x)
45
+ return x
46
+
47
+
48
+ class MLP(nn.Module):
49
+ def __init__(self, width):
50
+ super().__init__()
51
+ self.width = width
52
+ self.c_fc = nn.Linear(width, width * 4)
53
+ self.c_proj = nn.Linear(width * 4, width)
54
+ self.gelu = nn.GELU()
55
+
56
+ def forward(self, x):
57
+ return self.c_proj(self.gelu(self.c_fc(x)))
58
+
59
+
60
+ class QKVMultiheadAttention(nn.Module):
61
+ def __init__(self, n_heads: int, n_ctx: int):
62
+ super().__init__()
63
+ self.n_heads = n_heads
64
+ self.n_ctx = n_ctx
65
+
66
+ def forward(self, qkv):
67
+ bs, n_ctx, width = qkv.shape
68
+ attn_ch = width // self.n_heads // 3
69
+ scale = 1 / math.sqrt(math.sqrt(attn_ch))
70
+ qkv = qkv.view(bs, n_ctx, self.n_heads, -1)
71
+ q, k, v = th.split(qkv, attn_ch, dim=-1)
72
+ weight = th.einsum(
73
+ "bthc,bshc->bhts", q * scale, k * scale
74
+ ) # More stable with f16 than dividing afterwards
75
+ wdtype = weight.dtype
76
+ weight = th.softmax(weight.float(), dim=-1).type(wdtype)
77
+ return th.einsum("bhts,bshc->bthc", weight, v).reshape(bs, n_ctx, -1)
78
+
79
+
80
+ class ResidualAttentionBlock(nn.Module):
81
+ def __init__(
82
+ self,
83
+ n_ctx: int,
84
+ width: int,
85
+ heads: int,
86
+ ):
87
+ super().__init__()
88
+
89
+ self.attn = MultiheadAttention(
90
+ n_ctx,
91
+ width,
92
+ heads,
93
+ )
94
+ self.ln_1 = LayerNorm(width)
95
+ self.mlp = MLP(width)
96
+ self.ln_2 = LayerNorm(width)
97
+
98
+ def forward(self, x: th.Tensor):
99
+ x = x + self.attn(self.ln_1(x))
100
+ x = x + self.mlp(self.ln_2(x))
101
+ return x
102
+
103
+
104
+ class Transformer(nn.Module):
105
+ def __init__(
106
+ self,
107
+ n_ctx: int,
108
+ width: int,
109
+ layers: int,
110
+ heads: int,
111
+ ):
112
+ super().__init__()
113
+ self.n_ctx = n_ctx
114
+ self.width = width
115
+ self.layers = layers
116
+ self.resblocks = nn.ModuleList(
117
+ [
118
+ ResidualAttentionBlock(
119
+ n_ctx,
120
+ width,
121
+ heads,
122
+ )
123
+ for _ in range(layers)
124
+ ]
125
+ )
126
+
127
+ def forward(self, x: th.Tensor):
128
+ for block in self.resblocks:
129
+ x = block(x)
130
+ return x
model-card.md ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Overview
2
+
3
+ This card describes the diffusion model GLIDE (filtered) and noised CLIP model described in the paper [GLIDE: Towards
4
+ Photorealistic Image Generation and Editing with Text-Guided Diffusion Models](https://arxiv.org/abs/2112.10741)
5
+
6
+ # Datasets
7
+
8
+ GLIDE (filtered) was trained on a filtered version of a dataset comprised of several hundred million text-image pairs
9
+ collected from the internet. We constructed a set of filters intended to remove all images of people, violent objects, and some
10
+ and hate symbols (see Appendix F of the paper for details). The size of the dataset after filtering was approximately
11
+ 67M text-image pairs.
12
+
13
+ Our noised CLIP model which was trained on the dataset described above, augmented with a filtered version of the dataset used
14
+ to train the [original CLIP models](https://github.com/openai/clip). The total size of this augmented dataset is approximately 137M pairs.
15
+
16
+ # Performance
17
+
18
+ Qualitatively, we find that the generated images from GLIDE (filtered) often look semi-realistic, but the small size of the model hinders
19
+ its ability to bind attributes to objects and perform compositional tasks. Because the dataset used to train GLIDE
20
+ (filtered) has been preprocessed to remove images of people, this also limits its world knowledge, especially in regard
21
+ to concepts that involve people.
22
+ Finally, due to the dataset used to train GLIDE (filtered), the model has reduced capabilities to compose multiple objects in complex ways compared to models of a similar size trained on our internal dataset.
23
+
24
+ We do not directly measure quantitative metrics for GLIDE (filtered). In particular, most of the evaluations we report for our other models are biased against GLIDE (filtered), since they use prompts that often require generations of people. Evaluating people-free models remains an open area of research.
25
+
26
+ # Intended Use
27
+
28
+ We release these models to help advance research in generative modeling. Due to the limitations and biases of GLIDE (filtered), we do not currently recommend it for commercial use.
29
+
30
+ Functionally, these models are intended to be able to perform the following tasks for research purposes:
31
+ * Generate images from natural language prompts
32
+ * Iteratively edit and refine images using inpainting
33
+
34
+ These models are explicitly not intended to generate images of people or other subjects we filtered for (see Appendix F of the paper for details).
35
+
36
+ # Limitations
37
+
38
+ Despite the dataset filtering applied before training, GLIDE (filtered) continues to exhibit biases that extend beyond those found in images of people.
39
+ We explore some of these biases in our paper. For example:
40
+
41
+ * It produces different outputs when asked to generate toys for boys and toys for girls.
42
+ * It gravitates toward generating images of churches when asked to generate "a religious place",
43
+ and this bias is amplified by classifier-free guidance.
44
+ * It may have a greater propensity for generating hate symbols other than swastikas and confederate flags. Our filter
45
+ for hate symbols focused specifically on these two cases, as we found few relevant images of hate symbols in our
46
+ dataset. However, we also found that the model has diminished capabilities across a wider set of symbols.
47
+
48
+ GLIDE (filtered) can fail to produce realistic outputs for complex prompts or for prompts that involve concepts that are
49
+ not well-represented in its training data. While the data for the model was filtered to remove certain types of images,
50
+ the data still exhibits biases toward Western-centric concepts.
notebooks/clip_guided.ipynb ADDED
@@ -0,0 +1,246 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "cells": [
3
+ {
4
+ "cell_type": "code",
5
+ "execution_count": null,
6
+ "metadata": {},
7
+ "outputs": [],
8
+ "source": [
9
+ "# Run this line in Colab to install the package if it is\n",
10
+ "# not already installed.\n",
11
+ "!pip install git+https://github.com/openai/glide-text2im"
12
+ ]
13
+ },
14
+ {
15
+ "cell_type": "code",
16
+ "execution_count": null,
17
+ "metadata": {},
18
+ "outputs": [],
19
+ "source": [
20
+ "from PIL import Image\n",
21
+ "from IPython.display import display\n",
22
+ "import torch as th\n",
23
+ "import torch.nn as nn\n",
24
+ "\n",
25
+ "from glide_text2im.clip.model_creation import create_clip_model\n",
26
+ "from glide_text2im.download import load_checkpoint\n",
27
+ "from glide_text2im.model_creation import (\n",
28
+ " create_model_and_diffusion,\n",
29
+ " model_and_diffusion_defaults,\n",
30
+ " model_and_diffusion_defaults_upsampler,\n",
31
+ ")\n",
32
+ "from glide_text2im.tokenizer.simple_tokenizer import SimpleTokenizer"
33
+ ]
34
+ },
35
+ {
36
+ "cell_type": "code",
37
+ "execution_count": null,
38
+ "metadata": {},
39
+ "outputs": [],
40
+ "source": [
41
+ "# This notebook supports both CPU and GPU.\n",
42
+ "# On CPU, generating one sample may take on the order of 20 minutes.\n",
43
+ "# On a GPU, it should be under a minute.\n",
44
+ "\n",
45
+ "has_cuda = th.cuda.is_available()\n",
46
+ "device = th.device('cpu' if not has_cuda else 'cuda')"
47
+ ]
48
+ },
49
+ {
50
+ "cell_type": "code",
51
+ "execution_count": null,
52
+ "metadata": {},
53
+ "outputs": [],
54
+ "source": [
55
+ "# Create base model.\n",
56
+ "options = model_and_diffusion_defaults()\n",
57
+ "options['use_fp16'] = has_cuda\n",
58
+ "options['timestep_respacing'] = '100' # use 100 diffusion steps for fast sampling\n",
59
+ "model, diffusion = create_model_and_diffusion(**options)\n",
60
+ "model.eval()\n",
61
+ "if has_cuda:\n",
62
+ " model.convert_to_fp16()\n",
63
+ "model.to(device)\n",
64
+ "model.load_state_dict(load_checkpoint('base', device))\n",
65
+ "print('total base parameters', sum(x.numel() for x in model.parameters()))"
66
+ ]
67
+ },
68
+ {
69
+ "cell_type": "code",
70
+ "execution_count": null,
71
+ "metadata": {},
72
+ "outputs": [],
73
+ "source": [
74
+ "# Create upsampler model.\n",
75
+ "options_up = model_and_diffusion_defaults_upsampler()\n",
76
+ "options_up['use_fp16'] = has_cuda\n",
77
+ "options_up['timestep_respacing'] = 'fast27' # use 27 diffusion steps for very fast sampling\n",
78
+ "model_up, diffusion_up = create_model_and_diffusion(**options_up)\n",
79
+ "model_up.eval()\n",
80
+ "if has_cuda:\n",
81
+ " model_up.convert_to_fp16()\n",
82
+ "model_up.to(device)\n",
83
+ "model_up.load_state_dict(load_checkpoint('upsample', device))\n",
84
+ "print('total upsampler parameters', sum(x.numel() for x in model_up.parameters()))"
85
+ ]
86
+ },
87
+ {
88
+ "cell_type": "code",
89
+ "execution_count": null,
90
+ "metadata": {},
91
+ "outputs": [],
92
+ "source": [
93
+ "# Create CLIP model.\n",
94
+ "clip_model = create_clip_model(device=device)\n",
95
+ "clip_model.image_encoder.load_state_dict(load_checkpoint('clip/image-enc', device))\n",
96
+ "clip_model.text_encoder.load_state_dict(load_checkpoint('clip/text-enc', device))"
97
+ ]
98
+ },
99
+ {
100
+ "cell_type": "code",
101
+ "execution_count": null,
102
+ "metadata": {},
103
+ "outputs": [],
104
+ "source": [
105
+ "def show_images(batch: th.Tensor):\n",
106
+ " \"\"\" Display a batch of images inline. \"\"\"\n",
107
+ " scaled = ((batch + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu()\n",
108
+ " reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3])\n",
109
+ " display(Image.fromarray(reshaped.numpy()))"
110
+ ]
111
+ },
112
+ {
113
+ "cell_type": "code",
114
+ "execution_count": null,
115
+ "metadata": {},
116
+ "outputs": [],
117
+ "source": [
118
+ "# Sampling parameters\n",
119
+ "prompt = \"an oil painting of a corgi\"\n",
120
+ "batch_size = 1\n",
121
+ "guidance_scale = 3.0\n",
122
+ "\n",
123
+ "# Tune this parameter to control the sharpness of 256x256 images.\n",
124
+ "# A value of 1.0 is sharper, but sometimes results in grainy artifacts.\n",
125
+ "upsample_temp = 0.997"
126
+ ]
127
+ },
128
+ {
129
+ "cell_type": "code",
130
+ "execution_count": null,
131
+ "metadata": {},
132
+ "outputs": [],
133
+ "source": [
134
+ "##############################\n",
135
+ "# Sample from the base model #\n",
136
+ "##############################\n",
137
+ "\n",
138
+ "# Create the text tokens to feed to the model.\n",
139
+ "tokens = model.tokenizer.encode(prompt)\n",
140
+ "tokens, mask = model.tokenizer.padded_tokens_and_mask(\n",
141
+ " tokens, options['text_ctx']\n",
142
+ ")\n",
143
+ "\n",
144
+ "# Pack the tokens together into model kwargs.\n",
145
+ "model_kwargs = dict(\n",
146
+ " tokens=th.tensor([tokens] * batch_size, device=device),\n",
147
+ " mask=th.tensor([mask] * batch_size, dtype=th.bool, device=device),\n",
148
+ ")\n",
149
+ "\n",
150
+ "# Setup guidance function for CLIP model.\n",
151
+ "cond_fn = clip_model.cond_fn([prompt] * batch_size, guidance_scale)\n",
152
+ "\n",
153
+ "# Sample from the base model.\n",
154
+ "model.del_cache()\n",
155
+ "samples = diffusion.p_sample_loop(\n",
156
+ " model,\n",
157
+ " (batch_size, 3, options[\"image_size\"], options[\"image_size\"]),\n",
158
+ " device=device,\n",
159
+ " clip_denoised=True,\n",
160
+ " progress=True,\n",
161
+ " model_kwargs=model_kwargs,\n",
162
+ " cond_fn=cond_fn,\n",
163
+ ")\n",
164
+ "model.del_cache()\n",
165
+ "\n",
166
+ "# Show the output\n",
167
+ "show_images(samples)"
168
+ ]
169
+ },
170
+ {
171
+ "cell_type": "code",
172
+ "execution_count": null,
173
+ "metadata": {},
174
+ "outputs": [],
175
+ "source": [
176
+ "##############################\n",
177
+ "# Upsample the 64x64 samples #\n",
178
+ "##############################\n",
179
+ "\n",
180
+ "tokens = model_up.tokenizer.encode(prompt)\n",
181
+ "tokens, mask = model_up.tokenizer.padded_tokens_and_mask(\n",
182
+ " tokens, options_up['text_ctx']\n",
183
+ ")\n",
184
+ "\n",
185
+ "# Create the model conditioning dict.\n",
186
+ "model_kwargs = dict(\n",
187
+ " # Low-res image to upsample.\n",
188
+ " low_res=((samples+1)*127.5).round()/127.5 - 1,\n",
189
+ "\n",
190
+ " # Text tokens\n",
191
+ " tokens=th.tensor(\n",
192
+ " [tokens] * batch_size, device=device\n",
193
+ " ),\n",
194
+ " mask=th.tensor(\n",
195
+ " [mask] * batch_size,\n",
196
+ " dtype=th.bool,\n",
197
+ " device=device,\n",
198
+ " ),\n",
199
+ ")\n",
200
+ "\n",
201
+ "# Sample from the base model.\n",
202
+ "model_up.del_cache()\n",
203
+ "up_shape = (batch_size, 3, options_up[\"image_size\"], options_up[\"image_size\"])\n",
204
+ "up_samples = diffusion_up.ddim_sample_loop(\n",
205
+ " model_up,\n",
206
+ " up_shape,\n",
207
+ " noise=th.randn(up_shape, device=device) * upsample_temp,\n",
208
+ " device=device,\n",
209
+ " clip_denoised=True,\n",
210
+ " progress=True,\n",
211
+ " model_kwargs=model_kwargs,\n",
212
+ " cond_fn=None,\n",
213
+ ")[:batch_size]\n",
214
+ "model_up.del_cache()\n",
215
+ "\n",
216
+ "# Show the output\n",
217
+ "show_images(up_samples)"
218
+ ]
219
+ }
220
+ ],
221
+ "metadata": {
222
+ "interpreter": {
223
+ "hash": "e7d6e62d90e7e85f9a0faa7f0b1d576302d7ae6108e9fe361594f8e1c8b05781"
224
+ },
225
+ "kernelspec": {
226
+ "display_name": "Python 3",
227
+ "language": "python",
228
+ "name": "python3"
229
+ },
230
+ "language_info": {
231
+ "codemirror_mode": {
232
+ "name": "ipython",
233
+ "version": 3
234
+ },
235
+ "file_extension": ".py",
236
+ "mimetype": "text/x-python",
237
+ "name": "python",
238
+ "nbconvert_exporter": "python",
239
+ "pygments_lexer": "ipython3",
240
+ "version": "3.7.3"
241
+ },
242
+ "accelerator": "GPU"
243
+ },
244
+ "nbformat": 4,
245
+ "nbformat_minor": 2
246
+ }
notebooks/grass.png ADDED
notebooks/inpaint.ipynb ADDED
@@ -0,0 +1,302 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "cells": [
3
+ {
4
+ "cell_type": "code",
5
+ "execution_count": null,
6
+ "metadata": {},
7
+ "outputs": [],
8
+ "source": [
9
+ "# Run this line in Colab to install the package if it is\n",
10
+ "# not already installed.\n",
11
+ "!pip install git+https://github.com/openai/glide-text2im"
12
+ ]
13
+ },
14
+ {
15
+ "cell_type": "code",
16
+ "execution_count": null,
17
+ "metadata": {},
18
+ "outputs": [],
19
+ "source": [
20
+ "from typing import Tuple\n",
21
+ "\n",
22
+ "from IPython.display import display\n",
23
+ "from PIL import Image\n",
24
+ "import numpy as np\n",
25
+ "import torch as th\n",
26
+ "import torch.nn.functional as F\n",
27
+ "\n",
28
+ "from glide_text2im.download import load_checkpoint\n",
29
+ "from glide_text2im.model_creation import (\n",
30
+ " create_model_and_diffusion,\n",
31
+ " model_and_diffusion_defaults,\n",
32
+ " model_and_diffusion_defaults_upsampler\n",
33
+ ")"
34
+ ]
35
+ },
36
+ {
37
+ "cell_type": "code",
38
+ "execution_count": null,
39
+ "metadata": {},
40
+ "outputs": [],
41
+ "source": [
42
+ "# This notebook supports both CPU and GPU.\n",
43
+ "# On CPU, generating one sample may take on the order of 20 minutes.\n",
44
+ "# On a GPU, it should be under a minute.\n",
45
+ "\n",
46
+ "has_cuda = th.cuda.is_available()\n",
47
+ "device = th.device('cpu' if not has_cuda else 'cuda')"
48
+ ]
49
+ },
50
+ {
51
+ "cell_type": "code",
52
+ "execution_count": null,
53
+ "metadata": {},
54
+ "outputs": [],
55
+ "source": [
56
+ "# Create base model.\n",
57
+ "options = model_and_diffusion_defaults()\n",
58
+ "options['inpaint'] = True\n",
59
+ "options['use_fp16'] = has_cuda\n",
60
+ "options['timestep_respacing'] = '100' # use 100 diffusion steps for fast sampling\n",
61
+ "model, diffusion = create_model_and_diffusion(**options)\n",
62
+ "model.eval()\n",
63
+ "if has_cuda:\n",
64
+ " model.convert_to_fp16()\n",
65
+ "model.to(device)\n",
66
+ "model.load_state_dict(load_checkpoint('base-inpaint', device))\n",
67
+ "print('total base parameters', sum(x.numel() for x in model.parameters()))"
68
+ ]
69
+ },
70
+ {
71
+ "cell_type": "code",
72
+ "execution_count": null,
73
+ "metadata": {},
74
+ "outputs": [],
75
+ "source": [
76
+ "# Create upsampler model.\n",
77
+ "options_up = model_and_diffusion_defaults_upsampler()\n",
78
+ "options_up['inpaint'] = True\n",
79
+ "options_up['use_fp16'] = has_cuda\n",
80
+ "options_up['timestep_respacing'] = 'fast27' # use 27 diffusion steps for very fast sampling\n",
81
+ "model_up, diffusion_up = create_model_and_diffusion(**options_up)\n",
82
+ "model_up.eval()\n",
83
+ "if has_cuda:\n",
84
+ " model_up.convert_to_fp16()\n",
85
+ "model_up.to(device)\n",
86
+ "model_up.load_state_dict(load_checkpoint('upsample-inpaint', device))\n",
87
+ "print('total upsampler parameters', sum(x.numel() for x in model_up.parameters()))"
88
+ ]
89
+ },
90
+ {
91
+ "cell_type": "code",
92
+ "execution_count": null,
93
+ "metadata": {},
94
+ "outputs": [],
95
+ "source": [
96
+ "def show_images(batch: th.Tensor):\n",
97
+ " \"\"\" Display a batch of images inline. \"\"\"\n",
98
+ " scaled = ((batch + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu()\n",
99
+ " reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3])\n",
100
+ " display(Image.fromarray(reshaped.numpy()))\n",
101
+ "\n",
102
+ "def read_image(path: str, size: int = 256) -> Tuple[th.Tensor, th.Tensor]:\n",
103
+ " pil_img = Image.open(path).convert('RGB')\n",
104
+ " pil_img = pil_img.resize((size, size), resample=Image.BICUBIC)\n",
105
+ " img = np.array(pil_img)\n",
106
+ " return th.from_numpy(img)[None].permute(0, 3, 1, 2).float() / 127.5 - 1"
107
+ ]
108
+ },
109
+ {
110
+ "cell_type": "code",
111
+ "execution_count": null,
112
+ "metadata": {},
113
+ "outputs": [],
114
+ "source": [
115
+ "# Sampling parameters\n",
116
+ "prompt = \"a corgi in a field\"\n",
117
+ "batch_size = 1\n",
118
+ "guidance_scale = 5.0\n",
119
+ "\n",
120
+ "# Tune this parameter to control the sharpness of 256x256 images.\n",
121
+ "# A value of 1.0 is sharper, but sometimes results in grainy artifacts.\n",
122
+ "upsample_temp = 0.997\n",
123
+ "\n",
124
+ "# Source image we are inpainting\n",
125
+ "source_image_256 = read_image('grass.png', size=256)\n",
126
+ "source_image_64 = read_image('grass.png', size=64)\n",
127
+ "\n",
128
+ "# The mask should always be a boolean 64x64 mask, and then we\n",
129
+ "# can upsample it for the second stage.\n",
130
+ "source_mask_64 = th.ones_like(source_image_64)[:, :1]\n",
131
+ "source_mask_64[:, :, 20:] = 0\n",
132
+ "source_mask_256 = F.interpolate(source_mask_64, (256, 256), mode='nearest')\n",
133
+ "\n",
134
+ "# Visualize the image we are inpainting\n",
135
+ "show_images(source_image_256 * source_mask_256)"
136
+ ]
137
+ },
138
+ {
139
+ "cell_type": "code",
140
+ "execution_count": null,
141
+ "metadata": {},
142
+ "outputs": [],
143
+ "source": [
144
+ "##############################\n",
145
+ "# Sample from the base model #\n",
146
+ "##############################\n",
147
+ "\n",
148
+ "# Create the text tokens to feed to the model.\n",
149
+ "tokens = model.tokenizer.encode(prompt)\n",
150
+ "tokens, mask = model.tokenizer.padded_tokens_and_mask(\n",
151
+ " tokens, options['text_ctx']\n",
152
+ ")\n",
153
+ "\n",
154
+ "# Create the classifier-free guidance tokens (empty)\n",
155
+ "full_batch_size = batch_size * 2\n",
156
+ "uncond_tokens, uncond_mask = model.tokenizer.padded_tokens_and_mask(\n",
157
+ " [], options['text_ctx']\n",
158
+ ")\n",
159
+ "\n",
160
+ "# Pack the tokens together into model kwargs.\n",
161
+ "model_kwargs = dict(\n",
162
+ " tokens=th.tensor(\n",
163
+ " [tokens] * batch_size + [uncond_tokens] * batch_size, device=device\n",
164
+ " ),\n",
165
+ " mask=th.tensor(\n",
166
+ " [mask] * batch_size + [uncond_mask] * batch_size,\n",
167
+ " dtype=th.bool,\n",
168
+ " device=device,\n",
169
+ " ),\n",
170
+ "\n",
171
+ " # Masked inpainting image\n",
172
+ " inpaint_image=(source_image_64 * source_mask_64).repeat(full_batch_size, 1, 1, 1).to(device),\n",
173
+ " inpaint_mask=source_mask_64.repeat(full_batch_size, 1, 1, 1).to(device),\n",
174
+ ")\n",
175
+ "\n",
176
+ "# Create an classifier-free guidance sampling function\n",
177
+ "def model_fn(x_t, ts, **kwargs):\n",
178
+ " half = x_t[: len(x_t) // 2]\n",
179
+ " combined = th.cat([half, half], dim=0)\n",
180
+ " model_out = model(combined, ts, **kwargs)\n",
181
+ " eps, rest = model_out[:, :3], model_out[:, 3:]\n",
182
+ " cond_eps, uncond_eps = th.split(eps, len(eps) // 2, dim=0)\n",
183
+ " half_eps = uncond_eps + guidance_scale * (cond_eps - uncond_eps)\n",
184
+ " eps = th.cat([half_eps, half_eps], dim=0)\n",
185
+ " return th.cat([eps, rest], dim=1)\n",
186
+ "\n",
187
+ "def denoised_fn(x_start):\n",
188
+ " # Force the model to have the exact right x_start predictions\n",
189
+ " # for the part of the image which is known.\n",
190
+ " return (\n",
191
+ " x_start * (1 - model_kwargs['inpaint_mask'])\n",
192
+ " + model_kwargs['inpaint_image'] * model_kwargs['inpaint_mask']\n",
193
+ " )\n",
194
+ "\n",
195
+ "# Sample from the base model.\n",
196
+ "model.del_cache()\n",
197
+ "samples = diffusion.p_sample_loop(\n",
198
+ " model_fn,\n",
199
+ " (full_batch_size, 3, options[\"image_size\"], options[\"image_size\"]),\n",
200
+ " device=device,\n",
201
+ " clip_denoised=True,\n",
202
+ " progress=True,\n",
203
+ " model_kwargs=model_kwargs,\n",
204
+ " cond_fn=None,\n",
205
+ " denoised_fn=denoised_fn,\n",
206
+ ")[:batch_size]\n",
207
+ "model.del_cache()\n",
208
+ "\n",
209
+ "# Show the output\n",
210
+ "show_images(samples)"
211
+ ]
212
+ },
213
+ {
214
+ "cell_type": "code",
215
+ "execution_count": null,
216
+ "metadata": {},
217
+ "outputs": [],
218
+ "source": [
219
+ "##############################\n",
220
+ "# Upsample the 64x64 samples #\n",
221
+ "##############################\n",
222
+ "\n",
223
+ "tokens = model_up.tokenizer.encode(prompt)\n",
224
+ "tokens, mask = model_up.tokenizer.padded_tokens_and_mask(\n",
225
+ " tokens, options_up['text_ctx']\n",
226
+ ")\n",
227
+ "\n",
228
+ "# Create the model conditioning dict.\n",
229
+ "model_kwargs = dict(\n",
230
+ " # Low-res image to upsample.\n",
231
+ " low_res=((samples+1)*127.5).round()/127.5 - 1,\n",
232
+ "\n",
233
+ " # Text tokens\n",
234
+ " tokens=th.tensor(\n",
235
+ " [tokens] * batch_size, device=device\n",
236
+ " ),\n",
237
+ " mask=th.tensor(\n",
238
+ " [mask] * batch_size,\n",
239
+ " dtype=th.bool,\n",
240
+ " device=device,\n",
241
+ " ),\n",
242
+ "\n",
243
+ " # Masked inpainting image.\n",
244
+ " inpaint_image=(source_image_256 * source_mask_256).repeat(batch_size, 1, 1, 1).to(device),\n",
245
+ " inpaint_mask=source_mask_256.repeat(batch_size, 1, 1, 1).to(device),\n",
246
+ ")\n",
247
+ "\n",
248
+ "def denoised_fn(x_start):\n",
249
+ " # Force the model to have the exact right x_start predictions\n",
250
+ " # for the part of the image which is known.\n",
251
+ " return (\n",
252
+ " x_start * (1 - model_kwargs['inpaint_mask'])\n",
253
+ " + model_kwargs['inpaint_image'] * model_kwargs['inpaint_mask']\n",
254
+ " )\n",
255
+ "\n",
256
+ "# Sample from the base model.\n",
257
+ "model_up.del_cache()\n",
258
+ "up_shape = (batch_size, 3, options_up[\"image_size\"], options_up[\"image_size\"])\n",
259
+ "up_samples = diffusion_up.p_sample_loop(\n",
260
+ " model_up,\n",
261
+ " up_shape,\n",
262
+ " noise=th.randn(up_shape, device=device) * upsample_temp,\n",
263
+ " device=device,\n",
264
+ " clip_denoised=True,\n",
265
+ " progress=True,\n",
266
+ " model_kwargs=model_kwargs,\n",
267
+ " cond_fn=None,\n",
268
+ " denoised_fn=denoised_fn,\n",
269
+ ")[:batch_size]\n",
270
+ "model_up.del_cache()\n",
271
+ "\n",
272
+ "# Show the output\n",
273
+ "show_images(up_samples)"
274
+ ]
275
+ }
276
+ ],
277
+ "metadata": {
278
+ "interpreter": {
279
+ "hash": "e7d6e62d90e7e85f9a0faa7f0b1d576302d7ae6108e9fe361594f8e1c8b05781"
280
+ },
281
+ "kernelspec": {
282
+ "display_name": "Python 3",
283
+ "language": "python",
284
+ "name": "python3"
285
+ },
286
+ "language_info": {
287
+ "codemirror_mode": {
288
+ "name": "ipython",
289
+ "version": 3
290
+ },
291
+ "file_extension": ".py",
292
+ "mimetype": "text/x-python",
293
+ "name": "python",
294
+ "nbconvert_exporter": "python",
295
+ "pygments_lexer": "ipython3",
296
+ "version": "3.7.3"
297
+ },
298
+ "accelerator": "GPU"
299
+ },
300
+ "nbformat": 4,
301
+ "nbformat_minor": 2
302
+ }
notebooks/text2im.ipynb ADDED
@@ -0,0 +1,251 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "cells": [
3
+ {
4
+ "cell_type": "code",
5
+ "execution_count": null,
6
+ "metadata": {},
7
+ "outputs": [],
8
+ "source": [
9
+ "# Run this line in Colab to install the package if it is\n",
10
+ "# not already installed.\n",
11
+ "!pip install git+https://github.com/openai/glide-text2im"
12
+ ]
13
+ },
14
+ {
15
+ "cell_type": "code",
16
+ "execution_count": null,
17
+ "metadata": {},
18
+ "outputs": [],
19
+ "source": [
20
+ "from PIL import Image\n",
21
+ "from IPython.display import display\n",
22
+ "import torch as th\n",
23
+ "\n",
24
+ "from glide_text2im.download import load_checkpoint\n",
25
+ "from glide_text2im.model_creation import (\n",
26
+ " create_model_and_diffusion,\n",
27
+ " model_and_diffusion_defaults,\n",
28
+ " model_and_diffusion_defaults_upsampler\n",
29
+ ")"
30
+ ]
31
+ },
32
+ {
33
+ "cell_type": "code",
34
+ "execution_count": null,
35
+ "metadata": {},
36
+ "outputs": [],
37
+ "source": [
38
+ "# This notebook supports both CPU and GPU.\n",
39
+ "# On CPU, generating one sample may take on the order of 20 minutes.\n",
40
+ "# On a GPU, it should be under a minute.\n",
41
+ "\n",
42
+ "has_cuda = th.cuda.is_available()\n",
43
+ "device = th.device('cpu' if not has_cuda else 'cuda')"
44
+ ]
45
+ },
46
+ {
47
+ "cell_type": "code",
48
+ "execution_count": null,
49
+ "metadata": {},
50
+ "outputs": [],
51
+ "source": [
52
+ "# Create base model.\n",
53
+ "options = model_and_diffusion_defaults()\n",
54
+ "options['use_fp16'] = has_cuda\n",
55
+ "options['timestep_respacing'] = '100' # use 100 diffusion steps for fast sampling\n",
56
+ "model, diffusion = create_model_and_diffusion(**options)\n",
57
+ "model.eval()\n",
58
+ "if has_cuda:\n",
59
+ " model.convert_to_fp16()\n",
60
+ "model.to(device)\n",
61
+ "model.load_state_dict(load_checkpoint('base', device))\n",
62
+ "print('total base parameters', sum(x.numel() for x in model.parameters()))"
63
+ ]
64
+ },
65
+ {
66
+ "cell_type": "code",
67
+ "execution_count": null,
68
+ "metadata": {},
69
+ "outputs": [],
70
+ "source": [
71
+ "# Create upsampler model.\n",
72
+ "options_up = model_and_diffusion_defaults_upsampler()\n",
73
+ "options_up['use_fp16'] = has_cuda\n",
74
+ "options_up['timestep_respacing'] = 'fast27' # use 27 diffusion steps for very fast sampling\n",
75
+ "model_up, diffusion_up = create_model_and_diffusion(**options_up)\n",
76
+ "model_up.eval()\n",
77
+ "if has_cuda:\n",
78
+ " model_up.convert_to_fp16()\n",
79
+ "model_up.to(device)\n",
80
+ "model_up.load_state_dict(load_checkpoint('upsample', device))\n",
81
+ "print('total upsampler parameters', sum(x.numel() for x in model_up.parameters()))"
82
+ ]
83
+ },
84
+ {
85
+ "cell_type": "code",
86
+ "execution_count": null,
87
+ "metadata": {},
88
+ "outputs": [],
89
+ "source": [
90
+ "def show_images(batch: th.Tensor):\n",
91
+ " \"\"\" Display a batch of images inline. \"\"\"\n",
92
+ " scaled = ((batch + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu()\n",
93
+ " reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3])\n",
94
+ " display(Image.fromarray(reshaped.numpy()))"
95
+ ]
96
+ },
97
+ {
98
+ "cell_type": "code",
99
+ "execution_count": null,
100
+ "metadata": {},
101
+ "outputs": [],
102
+ "source": [
103
+ "# Sampling parameters\n",
104
+ "prompt = \"an oil painting of a corgi\"\n",
105
+ "batch_size = 1\n",
106
+ "guidance_scale = 3.0\n",
107
+ "\n",
108
+ "# Tune this parameter to control the sharpness of 256x256 images.\n",
109
+ "# A value of 1.0 is sharper, but sometimes results in grainy artifacts.\n",
110
+ "upsample_temp = 0.997"
111
+ ]
112
+ },
113
+ {
114
+ "cell_type": "code",
115
+ "execution_count": null,
116
+ "metadata": {},
117
+ "outputs": [],
118
+ "source": [
119
+ "##############################\n",
120
+ "# Sample from the base model #\n",
121
+ "##############################\n",
122
+ "\n",
123
+ "# Create the text tokens to feed to the model.\n",
124
+ "tokens = model.tokenizer.encode(prompt)\n",
125
+ "tokens, mask = model.tokenizer.padded_tokens_and_mask(\n",
126
+ " tokens, options['text_ctx']\n",
127
+ ")\n",
128
+ "\n",
129
+ "# Create the classifier-free guidance tokens (empty)\n",
130
+ "full_batch_size = batch_size * 2\n",
131
+ "uncond_tokens, uncond_mask = model.tokenizer.padded_tokens_and_mask(\n",
132
+ " [], options['text_ctx']\n",
133
+ ")\n",
134
+ "\n",
135
+ "# Pack the tokens together into model kwargs.\n",
136
+ "model_kwargs = dict(\n",
137
+ " tokens=th.tensor(\n",
138
+ " [tokens] * batch_size + [uncond_tokens] * batch_size, device=device\n",
139
+ " ),\n",
140
+ " mask=th.tensor(\n",
141
+ " [mask] * batch_size + [uncond_mask] * batch_size,\n",
142
+ " dtype=th.bool,\n",
143
+ " device=device,\n",
144
+ " ),\n",
145
+ ")\n",
146
+ "\n",
147
+ "# Create a classifier-free guidance sampling function\n",
148
+ "def model_fn(x_t, ts, **kwargs):\n",
149
+ " half = x_t[: len(x_t) // 2]\n",
150
+ " combined = th.cat([half, half], dim=0)\n",
151
+ " model_out = model(combined, ts, **kwargs)\n",
152
+ " eps, rest = model_out[:, :3], model_out[:, 3:]\n",
153
+ " cond_eps, uncond_eps = th.split(eps, len(eps) // 2, dim=0)\n",
154
+ " half_eps = uncond_eps + guidance_scale * (cond_eps - uncond_eps)\n",
155
+ " eps = th.cat([half_eps, half_eps], dim=0)\n",
156
+ " return th.cat([eps, rest], dim=1)\n",
157
+ "\n",
158
+ "# Sample from the base model.\n",
159
+ "model.del_cache()\n",
160
+ "samples = diffusion.p_sample_loop(\n",
161
+ " model_fn,\n",
162
+ " (full_batch_size, 3, options[\"image_size\"], options[\"image_size\"]),\n",
163
+ " device=device,\n",
164
+ " clip_denoised=True,\n",
165
+ " progress=True,\n",
166
+ " model_kwargs=model_kwargs,\n",
167
+ " cond_fn=None,\n",
168
+ ")[:batch_size]\n",
169
+ "model.del_cache()\n",
170
+ "\n",
171
+ "# Show the output\n",
172
+ "show_images(samples)"
173
+ ]
174
+ },
175
+ {
176
+ "cell_type": "code",
177
+ "execution_count": null,
178
+ "metadata": {},
179
+ "outputs": [],
180
+ "source": [
181
+ "##############################\n",
182
+ "# Upsample the 64x64 samples #\n",
183
+ "##############################\n",
184
+ "\n",
185
+ "tokens = model_up.tokenizer.encode(prompt)\n",
186
+ "tokens, mask = model_up.tokenizer.padded_tokens_and_mask(\n",
187
+ " tokens, options_up['text_ctx']\n",
188
+ ")\n",
189
+ "\n",
190
+ "# Create the model conditioning dict.\n",
191
+ "model_kwargs = dict(\n",
192
+ " # Low-res image to upsample.\n",
193
+ " low_res=((samples+1)*127.5).round()/127.5 - 1,\n",
194
+ "\n",
195
+ " # Text tokens\n",
196
+ " tokens=th.tensor(\n",
197
+ " [tokens] * batch_size, device=device\n",
198
+ " ),\n",
199
+ " mask=th.tensor(\n",
200
+ " [mask] * batch_size,\n",
201
+ " dtype=th.bool,\n",
202
+ " device=device,\n",
203
+ " ),\n",
204
+ ")\n",
205
+ "\n",
206
+ "# Sample from the base model.\n",
207
+ "model_up.del_cache()\n",
208
+ "up_shape = (batch_size, 3, options_up[\"image_size\"], options_up[\"image_size\"])\n",
209
+ "up_samples = diffusion_up.ddim_sample_loop(\n",
210
+ " model_up,\n",
211
+ " up_shape,\n",
212
+ " noise=th.randn(up_shape, device=device) * upsample_temp,\n",
213
+ " device=device,\n",
214
+ " clip_denoised=True,\n",
215
+ " progress=True,\n",
216
+ " model_kwargs=model_kwargs,\n",
217
+ " cond_fn=None,\n",
218
+ ")[:batch_size]\n",
219
+ "model_up.del_cache()\n",
220
+ "\n",
221
+ "# Show the output\n",
222
+ "show_images(up_samples)"
223
+ ]
224
+ }
225
+ ],
226
+ "metadata": {
227
+ "interpreter": {
228
+ "hash": "e7d6e62d90e7e85f9a0faa7f0b1d576302d7ae6108e9fe361594f8e1c8b05781"
229
+ },
230
+ "kernelspec": {
231
+ "display_name": "Python 3",
232
+ "language": "python",
233
+ "name": "python3"
234
+ },
235
+ "language_info": {
236
+ "codemirror_mode": {
237
+ "name": "ipython",
238
+ "version": 3
239
+ },
240
+ "file_extension": ".py",
241
+ "mimetype": "text/x-python",
242
+ "name": "python",
243
+ "nbconvert_exporter": "python",
244
+ "pygments_lexer": "ipython3",
245
+ "version": "3.7.3"
246
+ },
247
+ "accelerator": "GPU"
248
+ },
249
+ "nbformat": 4,
250
+ "nbformat_minor": 2
251
+ }
requirements.txt ADDED
@@ -0,0 +1,4 @@
 
 
 
 
1
+ git+https://github.com/openai/glide-text2im.git
2
+ fastapi
3
+ uvicorn
4
+ regex
setup.py ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from setuptools import setup
2
+
3
+ setup(
4
+ name="glide-text2im",
5
+ packages=[
6
+ "glide_text2im",
7
+ "glide_text2im.clip",
8
+ "glide_text2im.tokenizer",
9
+ ],
10
+ package_data={
11
+ "glide_text2im.tokenizer": [
12
+ "bpe_simple_vocab_16e6.txt.gz",
13
+ "encoder.json.gz",
14
+ "vocab.bpe.gz",
15
+ ],
16
+ "glide_text2im.clip": ["config.yaml"],
17
+ },
18
+ install_requires=[
19
+ "Pillow",
20
+ "attrs",
21
+ "torch",
22
+ "filelock",
23
+ "requests",
24
+ "tqdm",
25
+ "ftfy",
26
+ "regex",
27
+ ],
28
+ author="OpenAI",
29
+ )