Narsil HF staff commited on
Commit
647c28b
•
1 Parent(s): 621cfdb

Fp16 version.

Browse files
README.md CHANGED
@@ -6,16 +6,14 @@ tags:
6
  - stable-diffusion-diffusers
7
  - text-to-image
8
  extra_gated_prompt: |-
9
- One more step before getting this model.
10
- This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage.
11
- The CreativeML OpenRAIL License specifies:
12
-
13
  1. You can't use the model to deliberately produce nor share illegal or harmful outputs or content
14
- 2. CompVis claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license
15
  3. You may re-distribute the weights and use the model commercially and/or as a service. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully)
16
  Please read the full license here: https://huggingface.co/spaces/CompVis/stable-diffusion-license
17
-
18
- By clicking on "Access repository" below, you accept that your *contact information* (email address and username) can be shared with the model authors as well.
19
 
20
  extra_gated_fields:
21
  I have read the License and agree with its terms: checkbox
@@ -24,12 +22,10 @@ extra_gated_fields:
24
  # Stable Diffusion v1-4 Model Card
25
 
26
  Stable Diffusion is a latent text-to-image diffusion model capable of generating photo-realistic images given any text input.
27
- For more information about how Stable Diffusion functions, please have a look at [🤗's Stable Diffusion with 🧨Diffusers blog](https://huggingface.co/blog/stable_diffusion).
28
 
29
- The **Stable-Diffusion-v1-4** checkpoint was initialized with the weights of the [Stable-Diffusion-v1-2](https:/steps/huggingface.co/CompVis/stable-diffusion-v1-2)
30
- checkpoint and subsequently fine-tuned on 225k steps at resolution 512x512 on "laion-aesthetics v2 5+" and 10% dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598).
31
-
32
- This weights here are intended to be used with the 🧨 Diffusers library. If you are looking for the weights to be loaded into the CompVis Stable Diffusion codebase, [come here](https://huggingface.co/CompVis/stable-diffusion-v-1-4-original)
33
 
34
  ## Model Details
35
  - **Developed by:** Robin Rombach, Patrick Esser
@@ -53,8 +49,6 @@ This weights here are intended to be used with the 🧨 Diffusers library. If yo
53
 
54
  We recommend using [🤗's Diffusers library](https://github.com/huggingface/diffusers) to run Stable Diffusion.
55
 
56
- ### PyTorch
57
-
58
  ```bash
59
  pip install --upgrade diffusers transformers scipy
60
  ```
@@ -65,8 +59,7 @@ Run this command to log in with your HF Hub token if you haven't before:
65
  huggingface-cli login
66
  ```
67
 
68
- Running the pipeline with the default PNDM scheduler:
69
-
70
  ```python
71
  import torch
72
  from torch import autocast
@@ -75,32 +68,15 @@ from diffusers import StableDiffusionPipeline
75
  model_id = "CompVis/stable-diffusion-v1-4"
76
  device = "cuda"
77
 
78
-
79
  pipe = StableDiffusionPipeline.from_pretrained(model_id, use_auth_token=True)
80
  pipe = pipe.to(device)
81
 
82
- prompt = "a photo of an astronaut riding a horse on mars"
83
  with autocast("cuda"):
84
- image = pipe(prompt, guidance_scale=7.5).images[0]
85
 
86
- image.save("astronaut_rides_horse.png")
87
- ```
88
-
89
- **Note**:
90
- If you are limited by GPU memory and have less than 10GB of GPU RAM available, please make sure to load the StableDiffusionPipeline in float16 precision instead of the default float32 precision as done above. You can do so by telling diffusers to expect the weights to be in float16 precision:
91
-
92
-
93
- ```py
94
- import torch
95
-
96
- pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16, revision="fp16", use_auth_token=True)
97
- pipe = pipe.to(device)
98
-
99
- prompt = "a photo of an astronaut riding a horse on mars"
100
- with autocast("cuda"):
101
- image = pipe(prompt, guidance_scale=7.5).images[0]
102
-
103
- image.save("astronaut_rides_horse.png")
104
  ```
105
 
106
  To swap out the noise scheduler, pass it to `from_pretrained`:
@@ -113,81 +89,6 @@ model_id = "CompVis/stable-diffusion-v1-4"
113
  scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
114
  pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, use_auth_token=True)
115
  pipe = pipe.to("cuda")
116
-
117
- prompt = "a photo of an astronaut riding a horse on mars"
118
- with autocast("cuda"):
119
- image = pipe(prompt, guidance_scale=7.5).images[0]
120
-
121
- image.save("astronaut_rides_horse.png")
122
- ```
123
-
124
- ### JAX/Flax
125
-
126
- To use StableDiffusion on TPUs and GPUs for faster inference you can leverage JAX/Flax.
127
-
128
- Running the pipeline with default PNDMScheduler
129
-
130
- ```python
131
- import jax
132
- import numpy as np
133
- from flax.jax_utils import replicate
134
- from flax.training.common_utils import shard
135
-
136
- from diffusers import FlaxStableDiffusionPipeline
137
-
138
- pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
139
- "CompVis/stable-diffusion-v1-4", revision="flax", dtype=jax.numpy.bfloat16
140
- )
141
-
142
- prompt = "a photo of an astronaut riding a horse on mars"
143
-
144
- prng_seed = jax.random.PRNGKey(0)
145
- num_inference_steps = 50
146
-
147
- num_samples = jax.device_count()
148
- prompt = num_samples * [prompt]
149
- prompt_ids = pipeline.prepare_inputs(prompt)
150
-
151
- # shard inputs and rng
152
- params = replicate(params)
153
- prng_seed = jax.random.split(prng_seed, 8)
154
- prompt_ids = shard(prompt_ids)
155
-
156
- images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
157
- images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
158
- ```
159
-
160
- **Note**:
161
- If you are limited by TPU memory, please make sure to load the `FlaxStableDiffusionPipeline` in `bfloat16` precision instead of the default `float32` precision as done above. You can do so by telling diffusers to load the weights from "bf16" branch.
162
-
163
- ```python
164
- import jax
165
- import numpy as np
166
- from flax.jax_utils import replicate
167
- from flax.training.common_utils import shard
168
-
169
- from diffusers import FlaxStableDiffusionPipeline
170
-
171
- pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
172
- "CompVis/stable-diffusion-v1-4", revision="bf16", dtype=jax.numpy.bfloat16
173
- )
174
-
175
- prompt = "a photo of an astronaut riding a horse on mars"
176
-
177
- prng_seed = jax.random.PRNGKey(0)
178
- num_inference_steps = 50
179
-
180
- num_samples = jax.device_count()
181
- prompt = num_samples * [prompt]
182
- prompt_ids = pipeline.prepare_inputs(prompt)
183
-
184
- # shard inputs and rng
185
- params = replicate(params)
186
- prng_seed = jax.random.split(prng_seed, 8)
187
- prompt_ids = shard(prompt_ids)
188
-
189
- images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
190
- images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
191
  ```
192
 
193
  # Uses
@@ -239,8 +140,6 @@ Using the model to generate content that is cruel to individuals is a misuse of
239
  [LAION-5B](https://laion.ai/blog/laion-5b/) which contains adult material
240
  and is not fit for product use without additional safety mechanisms and
241
  considerations.
242
- - No additional measures were used to deduplicate the dataset. As a result, we observe some degree of memorization for images that are duplicated in the training data.
243
- The training data can be searched at [https://rom1504.github.io/clip-retrieval/](https://rom1504.github.io/clip-retrieval/) to possibly assist in the detection of memorized images.
244
 
245
  ### Bias
246
 
@@ -251,14 +150,6 @@ Texts and images from communities and cultures that use other languages are like
251
  This affects the overall output of the model, as white and western cultures are often set as the default. Further, the
252
  ability of the model to generate content with non-English prompts is significantly worse than with English-language prompts.
253
 
254
- ### Safety Module
255
-
256
- The intended use of this model is with the [Safety Checker](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py) in Diffusers.
257
- This checker works by checking model outputs against known hard-coded NSFW concepts.
258
- The concepts are intentionally hidden to reduce the likelihood of reverse-engineering this filter.
259
- Specifically, the checker compares the class probability of harmful concepts in the embedding space of the `CLIPTextModel` *after generation* of the images.
260
- The concepts are passed into the model with the generated image and compared to a hand-engineered weight for each NSFW concept.
261
-
262
 
263
  ## Training
264
 
@@ -281,8 +172,8 @@ We currently provide four checkpoints, which were trained as follows.
281
  - [`stable-diffusion-v1-2`](https://huggingface.co/CompVis/stable-diffusion-v1-2): Resumed from `stable-diffusion-v1-1`.
282
  515,000 steps at resolution `512x512` on "laion-improved-aesthetics" (a subset of laion2B-en,
283
  filtered to images with an original size `>= 512x512`, estimated aesthetics score `> 5.0`, and an estimated watermark probability `< 0.5`. The watermark estimate is from the LAION-5B metadata, the aesthetics score is estimated using an [improved aesthetics estimator](https://github.com/christophschuhmann/improved-aesthetic-predictor)).
284
- - [`stable-diffusion-v1-3`](https://huggingface.co/CompVis/stable-diffusion-v1-3): Resumed from `stable-diffusion-v1-2`. 195,000 steps at resolution `512x512` on "laion-improved-aesthetics" and 10 % dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598).
285
- - [`stable-diffusion-v1-4`](https://huggingface.co/CompVis/stable-diffusion-v1-4) Resumed from `stable-diffusion-v1-2`.225,000 steps at resolution `512x512` on "laion-aesthetics v2 5+" and 10 % dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598).
286
 
287
  - **Hardware:** 32 x 8 x A100 GPUs
288
  - **Optimizer:** AdamW
@@ -295,7 +186,7 @@ Evaluations with different classifier-free guidance scales (1.5, 2.0, 3.0, 4.0,
295
  5.0, 6.0, 7.0, 8.0) and 50 PLMS sampling
296
  steps show the relative improvements of the checkpoints:
297
 
298
- ![pareto](https://huggingface.co/CompVis/stable-diffusion/resolve/main/v1-variants-scores.jpg)
299
 
300
  Evaluated using 50 PLMS steps and 10000 random prompts from the COCO2017 validation set, evaluated at 512x512 resolution. Not optimized for FID scores.
301
  ## Environmental Impact
@@ -323,4 +214,4 @@ Based on that information, we estimate the following CO2 emissions using the [Ma
323
  }
324
  ```
325
 
326
- *This model card was written by: Robin Rombach and Patrick Esser and is based on the [DALL-E Mini model card](https://huggingface.co/dalle-mini/dalle-mini).*
 
6
  - stable-diffusion-diffusers
7
  - text-to-image
8
  extra_gated_prompt: |-
9
+ One more step before getting this model
10
+ This model is open access and available to all, but it has the CreativeML OpenRAIL-M license you have to be aware of before using it - don't worry you are just one click away!
11
+ By clicking on "Access repository" below, you accept that your *contact information* (email address and username) can be shared with the model authors as well.
12
+ Summary of the CreativeML OpenRAIL License:
13
  1. You can't use the model to deliberately produce nor share illegal or harmful outputs or content
14
+ 2. We claim no rights on the outputs you generate, you are free to use them and are accountable for their use which should not go against the provisions set in the license
15
  3. You may re-distribute the weights and use the model commercially and/or as a service. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully)
16
  Please read the full license here: https://huggingface.co/spaces/CompVis/stable-diffusion-license
 
 
17
 
18
  extra_gated_fields:
19
  I have read the License and agree with its terms: checkbox
 
22
  # Stable Diffusion v1-4 Model Card
23
 
24
  Stable Diffusion is a latent text-to-image diffusion model capable of generating photo-realistic images given any text input.
25
+ For more information about how Stable Diffusion functions, please have a look at [🤗's Stable Diffusion with D🧨iffusers blog](https://huggingface.co/blog/stable_diffusion).
26
 
27
+ The **Stable-Diffusion-v1-4** checkpoint was initialized with the weights of the [Stable-Diffusion-v1-3](https:/steps/huggingface.co/CompVis/stable-diffusion-v1-3)
28
+ checkpoint and subsequently fine-tuned on X steps on Y with Z.
 
 
29
 
30
  ## Model Details
31
  - **Developed by:** Robin Rombach, Patrick Esser
 
49
 
50
  We recommend using [🤗's Diffusers library](https://github.com/huggingface/diffusers) to run Stable Diffusion.
51
 
 
 
52
  ```bash
53
  pip install --upgrade diffusers transformers scipy
54
  ```
 
59
  huggingface-cli login
60
  ```
61
 
62
+ Running the pipeline with the default PLMS scheduler:
 
63
  ```python
64
  import torch
65
  from torch import autocast
 
68
  model_id = "CompVis/stable-diffusion-v1-4"
69
  device = "cuda"
70
 
71
+ generator = torch.Generator(device=device).manual_seed(0)
72
  pipe = StableDiffusionPipeline.from_pretrained(model_id, use_auth_token=True)
73
  pipe = pipe.to(device)
74
 
75
+ prompt = "a photograph of an astronaut riding a horse"
76
  with autocast("cuda"):
77
+ image = pipe(prompt, generator=generator)["sample"][0] # image here is in PIL format
78
 
79
+ image.save(f"astronaut_rides_horse.png")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
80
  ```
81
 
82
  To swap out the noise scheduler, pass it to `from_pretrained`:
 
89
  scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
90
  pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, use_auth_token=True)
91
  pipe = pipe.to("cuda")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
92
  ```
93
 
94
  # Uses
 
140
  [LAION-5B](https://laion.ai/blog/laion-5b/) which contains adult material
141
  and is not fit for product use without additional safety mechanisms and
142
  considerations.
 
 
143
 
144
  ### Bias
145
 
 
150
  This affects the overall output of the model, as white and western cultures are often set as the default. Further, the
151
  ability of the model to generate content with non-English prompts is significantly worse than with English-language prompts.
152
 
 
 
 
 
 
 
 
 
153
 
154
  ## Training
155
 
 
172
  - [`stable-diffusion-v1-2`](https://huggingface.co/CompVis/stable-diffusion-v1-2): Resumed from `stable-diffusion-v1-1`.
173
  515,000 steps at resolution `512x512` on "laion-improved-aesthetics" (a subset of laion2B-en,
174
  filtered to images with an original size `>= 512x512`, estimated aesthetics score `> 5.0`, and an estimated watermark probability `< 0.5`. The watermark estimate is from the LAION-5B metadata, the aesthetics score is estimated using an [improved aesthetics estimator](https://github.com/christophschuhmann/improved-aesthetic-predictor)).
175
+ - [`stable-diffusion-v1-3`](https://huggingface.co/CompVis/stable-diffusion-v1-3): Resumed from `stable-diffusion-v1-2`. 195,000 steps at resolution `512x512` on "laion-improved-aesthetics" and 10 % dropping of the text-conditioning to improve [classifier-free guidance sampling](https://arxiv.org/abs/2207.12598)
176
+ - [**`stable-diffusion-v1-4`**](https://huggingface.co/CompVis/stable-diffusion-v1-4) *To-fill-here*
177
 
178
  - **Hardware:** 32 x 8 x A100 GPUs
179
  - **Optimizer:** AdamW
 
186
  5.0, 6.0, 7.0, 8.0) and 50 PLMS sampling
187
  steps show the relative improvements of the checkpoints:
188
 
189
+ ![pareto](v1-variants-scores.jpg)
190
 
191
  Evaluated using 50 PLMS steps and 10000 random prompts from the COCO2017 validation set, evaluated at 512x512 resolution. Not optimized for FID scores.
192
  ## Environmental Impact
 
214
  }
215
  ```
216
 
217
+ *This model card was written by: Robin Rombach and Patrick Esser and is based on the [DALL-E Mini model card](https://huggingface.co/dalle-mini/dalle-mini).*
model_index.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "_class_name": "StableDiffusionPipeline",
3
- "_diffusers_version": "0.2.2",
4
  "feature_extractor": [
5
  "transformers",
6
  "CLIPFeatureExtractor"
 
1
  {
2
  "_class_name": "StableDiffusionPipeline",
3
+ "_diffusers_version": "0.2.3",
4
  "feature_extractor": [
5
  "transformers",
6
  "CLIPFeatureExtractor"
safety_checker/config.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "_name_or_path": "./safety_module",
3
  "architectures": [
4
  "StableDiffusionSafetyChecker"
5
  ],
@@ -68,6 +68,7 @@
68
  "sep_token_id": null,
69
  "task_specific_params": null,
70
  "temperature": 1.0,
 
71
  "tie_encoder_decoder": false,
72
  "tie_word_embeddings": true,
73
  "tokenizer_class": null,
@@ -75,7 +76,7 @@
75
  "top_p": 1.0,
76
  "torch_dtype": null,
77
  "torchscript": false,
78
- "transformers_version": "4.21.0.dev0",
79
  "typical_p": 1.0,
80
  "use_bfloat16": false,
81
  "vocab_size": 49408
@@ -86,7 +87,7 @@
86
  "num_attention_heads": 12,
87
  "num_hidden_layers": 12
88
  },
89
- "torch_dtype": "float32",
90
  "transformers_version": null,
91
  "vision_config": {
92
  "_name_or_path": "",
@@ -133,6 +134,7 @@
133
  "num_attention_heads": 16,
134
  "num_beam_groups": 1,
135
  "num_beams": 1,
 
136
  "num_hidden_layers": 24,
137
  "num_return_sequences": 1,
138
  "output_attentions": false,
@@ -150,6 +152,7 @@
150
  "sep_token_id": null,
151
  "task_specific_params": null,
152
  "temperature": 1.0,
 
153
  "tie_encoder_decoder": false,
154
  "tie_word_embeddings": true,
155
  "tokenizer_class": null,
@@ -157,7 +160,7 @@
157
  "top_p": 1.0,
158
  "torch_dtype": null,
159
  "torchscript": false,
160
- "transformers_version": "4.21.0.dev0",
161
  "typical_p": 1.0,
162
  "use_bfloat16": false
163
  },
 
1
  {
2
+ "_name_or_path": "./safety_checker",
3
  "architectures": [
4
  "StableDiffusionSafetyChecker"
5
  ],
 
68
  "sep_token_id": null,
69
  "task_specific_params": null,
70
  "temperature": 1.0,
71
+ "tf_legacy_loss": false,
72
  "tie_encoder_decoder": false,
73
  "tie_word_embeddings": true,
74
  "tokenizer_class": null,
 
76
  "top_p": 1.0,
77
  "torch_dtype": null,
78
  "torchscript": false,
79
+ "transformers_version": "4.21.1",
80
  "typical_p": 1.0,
81
  "use_bfloat16": false,
82
  "vocab_size": 49408
 
87
  "num_attention_heads": 12,
88
  "num_hidden_layers": 12
89
  },
90
+ "torch_dtype": "float16",
91
  "transformers_version": null,
92
  "vision_config": {
93
  "_name_or_path": "",
 
134
  "num_attention_heads": 16,
135
  "num_beam_groups": 1,
136
  "num_beams": 1,
137
+ "num_channels": 3,
138
  "num_hidden_layers": 24,
139
  "num_return_sequences": 1,
140
  "output_attentions": false,
 
152
  "sep_token_id": null,
153
  "task_specific_params": null,
154
  "temperature": 1.0,
155
+ "tf_legacy_loss": false,
156
  "tie_encoder_decoder": false,
157
  "tie_word_embeddings": true,
158
  "tokenizer_class": null,
 
160
  "top_p": 1.0,
161
  "torch_dtype": null,
162
  "torchscript": false,
163
+ "transformers_version": "4.21.1",
164
  "typical_p": 1.0,
165
  "use_bfloat16": false
166
  },
safety_checker/pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:193490b58ef62739077262e833bf091c66c29488058681ac25cf7df3d8190974
3
- size 1216061799
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d37ca6e57ace94e4c2f03ed0f67b6dc83e1ef1160892074917aa68b28e2afc1
3
+ size 608098599
scheduler/scheduler_config.json CHANGED
@@ -1,13 +1,9 @@
1
  {
2
  "_class_name": "PNDMScheduler",
3
- "_diffusers_version": "0.7.0.dev0",
4
  "beta_end": 0.012,
5
  "beta_schedule": "scaled_linear",
6
  "beta_start": 0.00085,
7
  "num_train_timesteps": 1000,
8
- "set_alpha_to_one": false,
9
- "skip_prk_steps": true,
10
- "steps_offset": 1,
11
- "trained_betas": null,
12
- "clip_sample": false
13
  }
 
1
  {
2
  "_class_name": "PNDMScheduler",
3
+ "_diffusers_version": "0.2.3",
4
  "beta_end": 0.012,
5
  "beta_schedule": "scaled_linear",
6
  "beta_start": 0.00085,
7
  "num_train_timesteps": 1000,
8
+ "skip_prk_steps": true
 
 
 
 
9
  }
text_encoder/config.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "_name_or_path": "openai/clip-vit-large-patch14",
3
  "architectures": [
4
  "CLIPTextModel"
5
  ],
@@ -18,7 +18,7 @@
18
  "num_attention_heads": 12,
19
  "num_hidden_layers": 12,
20
  "pad_token_id": 1,
21
- "torch_dtype": "float32",
22
- "transformers_version": "4.21.0.dev0",
23
  "vocab_size": 49408
24
  }
 
1
  {
2
+ "_name_or_path": "./text_encoder",
3
  "architectures": [
4
  "CLIPTextModel"
5
  ],
 
18
  "num_attention_heads": 12,
19
  "num_hidden_layers": 12,
20
  "pad_token_id": 1,
21
+ "torch_dtype": "float16",
22
+ "transformers_version": "4.21.1",
23
  "vocab_size": 49408
24
  }
text_encoder/pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:770a47a9ffdcfda0b05506a7888ed714d06131d60267e6cf52765d61cf59fd67
3
- size 492305335
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:88bd85efb0f84e70521633f578715afb2873db4f2615fdfb1f66e99934715865
3
+ size 246184375
tokenizer/tokenizer_config.json CHANGED
@@ -19,7 +19,7 @@
19
  },
20
  "errors": "replace",
21
  "model_max_length": 77,
22
- "name_or_path": "openai/clip-vit-large-patch14",
23
  "pad_token": "<|endoftext|>",
24
  "special_tokens_map_file": "./special_tokens_map.json",
25
  "tokenizer_class": "CLIPTokenizer",
 
19
  },
20
  "errors": "replace",
21
  "model_max_length": 77,
22
+ "name_or_path": "./tokenizer",
23
  "pad_token": "<|endoftext|>",
24
  "special_tokens_map_file": "./special_tokens_map.json",
25
  "tokenizer_class": "CLIPTokenizer",
unet/config.json CHANGED
@@ -1,6 +1,7 @@
1
  {
2
  "_class_name": "UNet2DConditionModel",
3
- "_diffusers_version": "0.2.2",
 
4
  "act_fn": "silu",
5
  "attention_head_dim": 8,
6
  "block_out_channels": [
 
1
  {
2
  "_class_name": "UNet2DConditionModel",
3
+ "_diffusers_version": "0.2.3",
4
+ "_name_or_path": "./unet",
5
  "act_fn": "silu",
6
  "attention_head_dim": 8,
7
  "block_out_channels": [
unet/diffusion_pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:62d48b4d841a3178511fa453df0dae59b22089ace64609cc9d5353d0a7f37c26
3
- size 3438354725
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d98edd280d5e040ee77f5802b8e3be3513de757335d1dedc4e495647e7c2d573
3
+ size 1719312805
vae/config.json CHANGED
@@ -1,6 +1,7 @@
1
  {
2
  "_class_name": "AutoencoderKL",
3
- "_diffusers_version": "0.2.2",
 
4
  "act_fn": "silu",
5
  "block_out_channels": [
6
  128,
 
1
  {
2
  "_class_name": "AutoencoderKL",
3
+ "_diffusers_version": "0.2.3",
4
+ "_name_or_path": "./vae",
5
  "act_fn": "silu",
6
  "block_out_channels": [
7
  128,
vae/diffusion_pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1b134cded8eb78b184aefb8805b6b572f36fa77b255c483665dda931fa0130c5
3
- size 334707217
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:51c8904bc921e1e6f354b5fa8e99a1c82ead2f0540114de21557b8abfbb24ad0
3
+ size 167399505