ProGamerGov commited on
Commit
caa7fd6
·
verified ·
1 Parent(s): 9a5450b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -1
README.md CHANGED
@@ -3,6 +3,7 @@ language:
3
  - en
4
  base_model:
5
  - Qwen/Qwen-Image
 
6
  pipeline_tag: text-to-image
7
  tags:
8
  - '360'
@@ -50,7 +51,10 @@ Training was first performed using nf4 quantization for 32 epochs (8 epochs coun
50
  Training then continued at int8 quantization for another 16 epochs (4 epochs counting the original + augmentations as a single epoch):
51
  - `qwen-360-diffusion-int8-bf16-v1.safetensors` was trained for a total of 48 epochs (2,304,000 steps)
52
 
53
- Training is currently being continued at int8 quantization for another set of epochs on the new [Qwen/Qwen-Image-2512](https://huggingface.co/Qwen/Qwen-Image-2512) base model.
 
 
 
54
 
55
  ---
56
 
 
3
  - en
4
  base_model:
5
  - Qwen/Qwen-Image
6
+ - Qwen/Qwen-Image-2512
7
  pipeline_tag: text-to-image
8
  tags:
9
  - '360'
 
51
  Training then continued at int8 quantization for another 16 epochs (4 epochs counting the original + augmentations as a single epoch):
52
  - `qwen-360-diffusion-int8-bf16-v1.safetensors` was trained for a total of 48 epochs (2,304,000 steps)
53
 
54
+ Training continued at int8 quantization for another x epochs on the [Qwen/Qwen-Image-2512](https://huggingface.co/Qwen/Qwen-Image-2512) base model:
55
+ - The total number of equirectangular images increased to 35k, and regularization images were removed from the dataset.
56
+ - The loss was adjusted to be based on the information density of equirectangular images.
57
+ - `qwen-360-diffusion-2512-int8-bf16-v2.safetensors` was trained for x epochs
58
 
59
  ---
60