Fix training details

README.md

@@ -35,16 +35,18 @@ The model is designed to be capable of producing equirectangular images that can
 
 ### Training Details
 
-Training was first performed using nf4 for 8 epochs.
-- `qwen-360-diffusion-int4-bf16-v1.safetensors` was trained for 7 epochs.
-- `qwen-360-diffusion-int4-bf16-v1-b.safetensors` was trained for 8 epochs.
+The training dataset consists of 32k unique 360 degree equirectangular images. Each image was randomly rotated horizontally 3 times for data augmentation (original + 3 rotations), providing a total of 128k training images. All 32k original 360 images were manually checked by humans for seams, polar artifacts, incorrect distortions, and other problems before their inclusion in the dataset.
 
-Training then continued at int8 for another 4 epochs.
-- `qwen-360-diffusion-int8-bf16-v1.safetensors` was trained for a total of 12 epochs.
+For regularization, 64k images were randomly selected from the [pexels-568k-internvl2](https://huggingface.co/datasets/CaptionEmporium/pexels-568k-internvl2) dataset and added to the training set.
 
-Each 360 degree training image was randomly rotated horizontally 3 times for data augmentation (original + 3 rotations).
+**Training timeline:** 3 months and 23 days
 
-For regularization, 64k images were randomly selected from the [pexels-568k-internvl2](https://huggingface.co/datasets/CaptionEmporium/pexels-568k-internvl2) dataset.
+Training was first performed using nf4 quantization for 32 epochs (8 epochs counting the original + augmentations as a single epoch):
+- `qwen-360-diffusion-int4-bf16-v1.safetensors` was trained for 28 epochs (1,344,000 steps)
+- `qwen-360-diffusion-int4-bf16-v1-b.safetensors` was trained for 32 epochs (1,536,000 steps)
+
+Training then continued at int8 quantization for another 16 epochs (4 epochs counting the original + augmentations as a single epoch):
+- `qwen-360-diffusion-int8-bf16-v1.safetensors` was trained for a total of 48 epochs (2,304,000 steps)
 
 ---