Update README.md
Browse files
README.md
CHANGED
|
@@ -40,6 +40,17 @@ mlabonne/orpo-dpo-mix-40k
|
|
| 40 |
|
| 41 |
ORPO
|
| 42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
|
| 45 |
#### Training Hyperparameters
|
|
@@ -48,6 +59,12 @@ ORPO
|
|
| 48 |
fp16 mixed precision
|
| 49 |
|
| 50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
|
| 52 |
## Evaluation
|
| 53 |
|
|
|
|
| 40 |
|
| 41 |
ORPO
|
| 42 |
|
| 43 |
+
### Training Parameters
|
| 44 |
+
## Training Arguments:
|
| 45 |
+
|
| 46 |
+
Learning Rate: 1e-5
|
| 47 |
+
Batch Size: 1
|
| 48 |
+
max_steps: 1
|
| 49 |
+
Block Size: 512
|
| 50 |
+
Warmup Ratio: 0.1
|
| 51 |
+
Weight Decay: 0.01
|
| 52 |
+
Gradient Accumulation: 4
|
| 53 |
+
Mixed Precision: bf16
|
| 54 |
|
| 55 |
|
| 56 |
#### Training Hyperparameters
|
|
|
|
| 59 |
fp16 mixed precision
|
| 60 |
|
| 61 |
|
| 62 |
+
### LoRA Configuration:
|
| 63 |
+
|
| 64 |
+
R: 16
|
| 65 |
+
Alpha: 32
|
| 66 |
+
Dropout: 0.05
|
| 67 |
+
|
| 68 |
|
| 69 |
## Evaluation
|
| 70 |
|