Added hyperparams to readme
Browse files
README.md
CHANGED
|
@@ -61,3 +61,18 @@ transcription = processor.batch_decode(predicted_ids)
|
|
| 61 |
|
| 62 |
# transcription: ['veliki broj poslovnih subjekata posluje sa minusom velik dio']
|
| 63 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
# transcription: ['veliki broj poslovnih subjekata posluje sa minusom velik dio']
|
| 63 |
```
|
| 64 |
+
|
| 65 |
+
## Training hyperparameters
|
| 66 |
+
|
| 67 |
+
In fine-tuning, the following arguments were used:
|
| 68 |
+
|
| 69 |
+
|arg | value|
|
| 70 |
+
|---|---|
|
| 71 |
+
|`group_by_length` |True |
|
| 72 |
+
| `per_device_train_batch_size`|16 |
|
| 73 |
+
|`gradient_accumulation_steps` |4 |
|
| 74 |
+
|`num_train_epochs` |8 |
|
| 75 |
+
|`gradient_checkpointing` |True |
|
| 76 |
+
|`fp16` |True |
|
| 77 |
+
|`learning_rate` | 3e-4|
|
| 78 |
+
|`warmup_steps` | 500|
|