updated README.MD
Browse files
README.md
CHANGED
|
@@ -49,8 +49,7 @@ target_modules: # Applied to all linear layers
|
|
| 49 |
```
|
| 50 |
|
| 51 |
### Training Hyperparameters
|
| 52 |
-
- **Epochs**:
|
| 53 |
-
- **Steps**: 550
|
| 54 |
- **Batch Size**: 2 per device (16 effective with gradient accumulation)
|
| 55 |
- **Learning Rate**: 5e-5 (cosine schedule)
|
| 56 |
- **Max Context**: 8,192 tokens
|
|
@@ -59,14 +58,14 @@ target_modules: # Applied to all linear layers
|
|
| 59 |
|
| 60 |
### Training Results
|
| 61 |
```
|
| 62 |
-
"final_train_loss": 0.
|
| 63 |
-
"final_eval_loss": 0.
|
| 64 |
-
"final_train_perplexity": 1.
|
| 65 |
-
"final_eval_perplexity": 1.
|
| 66 |
-
"final_token_accuracy": 0.
|
| 67 |
-
"initial_loss": 1.
|
| 68 |
-
"initial_perplexity": 5.
|
| 69 |
-
"initial_accuracy": 0.
|
| 70 |
```
|
| 71 |
|
| 72 |
## 🚀 Usage
|
|
|
|
| 49 |
```
|
| 50 |
|
| 51 |
### Training Hyperparameters
|
| 52 |
+
- **Epochs**: 3
|
|
|
|
| 53 |
- **Batch Size**: 2 per device (16 effective with gradient accumulation)
|
| 54 |
- **Learning Rate**: 5e-5 (cosine schedule)
|
| 55 |
- **Max Context**: 8,192 tokens
|
|
|
|
| 58 |
|
| 59 |
### Training Results
|
| 60 |
```
|
| 61 |
+
"final_train_loss": 0.2641,
|
| 62 |
+
"final_eval_loss": 0.37574875354766846,
|
| 63 |
+
"final_train_perplexity": 1.3022584156313823,
|
| 64 |
+
"final_eval_perplexity": 1.4560812525608204,
|
| 65 |
+
"final_token_accuracy": 0.9259863365441561,
|
| 66 |
+
"initial_loss": 1.6648,
|
| 67 |
+
"initial_perplexity": 5.284616220817229,
|
| 68 |
+
"initial_accuracy": 0.6015806214883923
|
| 69 |
```
|
| 70 |
|
| 71 |
## 🚀 Usage
|