patrickvonplaten commited on
Commit
96fe19f
1 Parent(s): 72b584c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +0 -45
README.md CHANGED
@@ -189,51 +189,6 @@ Based on that information, we estimate the following CO2 emissions using the [Ma
189
  - **Carbon Emitted (Power consumption x Time x Carbon produced based on location of power grid):** 11250 kg CO2 eq.
190
 
191
 
192
- ## Citation
193
-
194
- ```bibtex
195
- @InProceedings{Rombach_2022_CVPR,
196
- author = {Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bj\"orn},
197
- title = {High-Resolution Image Synthesis With Latent Diffusion Models},
198
- booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
199
- month = {June},
200
- year = {2022},
201
- pages = {10684-10695}
202
- }
203
- ```
204
-
205
- *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).*
206
- olution `512x512` on "laion-improved-aesthetics" (a subset of laion2B-en,
207
- 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)).
208
- - `sd-v1-3.ckpt`: Resumed from `sd-v1-2.ckpt`. 195k 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).
209
-
210
-
211
- - **Hardware:** 32 x 8 x A100 GPUs
212
- - **Optimizer:** AdamW
213
- - **Gradient Accumulations**: 2
214
- - **Batch:** 32 x 8 x 2 x 4 = 2048
215
- - **Learning rate:** warmup to 0.0001 for 10,000 steps and then kept constant
216
-
217
- ## Evaluation Results
218
- Evaluations with different classifier-free guidance scales (1.5, 2.0, 3.0, 4.0,
219
- 5.0, 6.0, 7.0, 8.0) and 50 PLMS sampling
220
- steps show the relative improvements of the checkpoints:
221
-
222
- ![pareto](v1-variants-scores.jpg)
223
-
224
- Evaluated using 50 PLMS steps and 10000 random prompts from the COCO2017 validation set, evaluated at 512x512 resolution. Not optimized for FID scores.
225
-
226
- ## Environmental Impact
227
-
228
- **Stable Diffusion v1** **Estimated Emissions**
229
- Based on that information, we estimate the following CO2 emissions using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700). The hardware, runtime, cloud provider, and compute region were utilized to estimate the carbon impact.
230
-
231
- - **Hardware Type:** A100 PCIe 40GB
232
- - **Hours used:** 150000
233
- - **Cloud Provider:** AWS
234
- - **Compute Region:** US-east
235
- - **Carbon Emitted (Power consumption x Time x Carbon produced based on location of power grid):** 11250 kg CO2 eq.
236
-
237
  ## Citation
238
 
239
  ```bibtex
 
189
  - **Carbon Emitted (Power consumption x Time x Carbon produced based on location of power grid):** 11250 kg CO2 eq.
190
 
191
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
192
  ## Citation
193
 
194
  ```bibtex