Update README.md
Browse files
README.md
CHANGED
|
@@ -18,8 +18,14 @@ from PIL import Image
|
|
| 18 |
### Training hyperparameters
|
| 19 |
|
| 20 |
|
| 21 |
-
|
| 22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
|
| 25 |
|
|
|
|
| 18 |
### Training hyperparameters
|
| 19 |
|
| 20 |
|
| 21 |
+
- Low-Rank Adaptation (LoRA)
|
| 22 |
+
- training = 7 epochs
|
| 23 |
+
- learning rate = 0.0002
|
| 24 |
+
- batch sizes = 1 per device
|
| 25 |
+
- gradient accumulation steps = 1.
|
| 26 |
+
- The optimizer "paged_adamw_32bit"
|
| 27 |
+
- 10 warmup steps
|
| 28 |
+
- 4-bit quantization with the NormalFloat 4 format
|
| 29 |
|
| 30 |
|
| 31 |
|