Commit
•
ed88b85
1
Parent(s):
25d5505
Update README.md
Browse files
README.md
CHANGED
|
@@ -23,3 +23,25 @@ This gemma model was trained 2x faster with [Unsloth](https://github.com/unsloth
|
|
| 23 |
|
| 24 |
# Evaluation
|
| 25 |
- **ViMMRC test set:** 0.8475 accuracy
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
# Evaluation
|
| 25 |
- **ViMMRC test set:** 0.8475 accuracy
|
| 26 |
+
|
| 27 |
+
### Training hyperparameters
|
| 28 |
+
|
| 29 |
+
The following hyperparameters were used during training:
|
| 30 |
+
- learning_rate: 0.0002
|
| 31 |
+
- train_batch_size: 16
|
| 32 |
+
- eval_batch_size: 16
|
| 33 |
+
- seed: 3407
|
| 34 |
+
- gradient_accumulation_steps: 4
|
| 35 |
+
- total_train_batch_size: 64
|
| 36 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 37 |
+
- lr_scheduler_type: cosine
|
| 38 |
+
- lr_scheduler_warmup_steps: 5
|
| 39 |
+
- num_epochs: 3
|
| 40 |
+
|
| 41 |
+
### Framework versions
|
| 42 |
+
|
| 43 |
+
- PEFT 0.10.0
|
| 44 |
+
- Transformers 4.40.2
|
| 45 |
+
- Pytorch 2.3.0
|
| 46 |
+
- Datasets 2.19.1
|
| 47 |
+
- Tokenizers 0.19.1
|