sayakpaul HF staff commited on
Commit
cd9b022
1 Parent(s): d173db6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -12
README.md CHANGED
@@ -91,22 +91,13 @@ Our training script was built on top of the official training script that we pro
91
  You can refer to [this script](https://github.com/patil-suraj/muse-experiments/blob/f71e7e79af24509ddb4e1b295a1d0ef8d8758dc9/ctrlnet/train_controlnet_webdataset.py) for full discolsure.
92
 
93
  #### Training data
94
- This checkpoint was first trained for 20,000 steps on LAION 6A resized to a max minimum dimension of 384.
95
- It was then further trained for 20,000 steps on laion 6a resized to a max minimum dimension of 1024 and
96
- then filtered to contain only minimum 1024 images. We found the further high resolution finetuning was
97
- necessary for image quality.
98
 
99
  #### Compute
100
- one 8xA100 machine
101
-
102
- #### Batch size
103
- Data parallel with a single gpu batch size of 8 for a total batch size of 64.
104
-
105
- #### Hyper Parameters
106
- Constant learning rate of 1e-4 scaled by batch size for total learning rate of 64e-4
107
 
108
  #### Mixed precision
109
- fp16
110
 
111
  #### Additional notes
112
 
 
91
  You can refer to [this script](https://github.com/patil-suraj/muse-experiments/blob/f71e7e79af24509ddb4e1b295a1d0ef8d8758dc9/ctrlnet/train_controlnet_webdataset.py) for full discolsure.
92
 
93
  #### Training data
94
+ The model was trained on 3M images from LAION aesthetic 6 plus subset, with batch size of 256 for 50k steps with constant learning rate of 3e-5.
 
 
 
95
 
96
  #### Compute
97
+ One 8xA100 machine
 
 
 
 
 
 
98
 
99
  #### Mixed precision
100
+ FP16
101
 
102
  #### Additional notes
103