ZeroCool94
commited on
Commit
•
f8d2192
1
Parent(s):
4a62b04
Update README.md
Browse files
README.md
CHANGED
@@ -87,9 +87,10 @@ image.save("fantasy_forest_illustration.png")
|
|
87 |
- #### Stable:
|
88 |
- [Sygil Diffusion v0.1](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.1.ckpt): Trained on Stable Diffusion 1.5 for 800,000 steps.
|
89 |
- [Sygil Diffusion v0.2](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.2.ckpt): Resumed from Sygil Diffusion v0.1 and trained for a total of 1.77 million steps.
|
90 |
-
- [Sygil Diffusion v0.3](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.3.ckpt): Resumed from Sygil Diffusion v0.2 and trained for a total of 2.01 million steps
|
|
|
91 |
- #### Beta:
|
92 |
-
- [sygil-diffusion-v0.
|
93 |
|
94 |
Note: Checkpoints under the Beta section are updated daily or at least 3-4 times a week. This is usually the equivalent of 1-2 training session,
|
95 |
this is done until they are stable enough to be moved into a proper release, usually every 1 or 2 weeks.
|
@@ -105,7 +106,7 @@ The model was trained on the following dataset:
|
|
105 |
|
106 |
**Hardware and others**
|
107 |
- **Hardware:** 1 x Nvidia RTX 3050 8GB GPU
|
108 |
-
- **Hours Trained:**
|
109 |
- **Optimizer:** AdamW
|
110 |
- **Adam Beta 1**: 0.9
|
111 |
- **Adam Beta 2**: 0.999
|
@@ -120,7 +121,7 @@ The model was trained on the following dataset:
|
|
120 |
- **Lora unet Learning Rate**: 1e-7
|
121 |
- **Lora Text Encoder Learning Rate**: 1e-7
|
122 |
- **Resolution**: 512 pixels
|
123 |
-
- **Total Training Steps:** 2,
|
124 |
|
125 |
|
126 |
Note: For the learning rate I'm testing something new, after changing from using the `constant` scheduler to `cosine_with_restarts` after v0.3 was released, I noticed
|
|
|
87 |
- #### Stable:
|
88 |
- [Sygil Diffusion v0.1](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.1.ckpt): Trained on Stable Diffusion 1.5 for 800,000 steps.
|
89 |
- [Sygil Diffusion v0.2](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.2.ckpt): Resumed from Sygil Diffusion v0.1 and trained for a total of 1.77 million steps.
|
90 |
+
- [Sygil Diffusion v0.3](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.3.ckpt): Resumed from Sygil Diffusion v0.2 and trained for a total of 2.01 million steps.
|
91 |
+
- [Sygil Diffusion v0.4](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.4.ckpt): Resumed from Sygil Diffusion v0.3 and trained for a total of 2.37 million steps.
|
92 |
- #### Beta:
|
93 |
+
- [sygil-diffusion-v0.4_2370200_lora.ckpt](https://huggingface.co/Sygil/Sygil-Diffusion/blob/main/sygil-diffusion-v0.4.ckpt): Resumed from Sygil Diffusion v0.3 and trained for a total of 2.37 million steps so far.
|
94 |
|
95 |
Note: Checkpoints under the Beta section are updated daily or at least 3-4 times a week. This is usually the equivalent of 1-2 training session,
|
96 |
this is done until they are stable enough to be moved into a proper release, usually every 1 or 2 weeks.
|
|
|
106 |
|
107 |
**Hardware and others**
|
108 |
- **Hardware:** 1 x Nvidia RTX 3050 8GB GPU
|
109 |
+
- **Hours Trained:** 857 hours approximately.
|
110 |
- **Optimizer:** AdamW
|
111 |
- **Adam Beta 1**: 0.9
|
112 |
- **Adam Beta 2**: 0.999
|
|
|
121 |
- **Lora unet Learning Rate**: 1e-7
|
122 |
- **Lora Text Encoder Learning Rate**: 1e-7
|
123 |
- **Resolution**: 512 pixels
|
124 |
+
- **Total Training Steps:** 2,370,200
|
125 |
|
126 |
|
127 |
Note: For the learning rate I'm testing something new, after changing from using the `constant` scheduler to `cosine_with_restarts` after v0.3 was released, I noticed
|