Upload hparams.yaml with huggingface_hub
Browse files- hparams.yaml +21 -0
hparams.yaml
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
nr_frozen_epochs: 0.3
|
| 2 |
+
keep_embeddings_frozen: true
|
| 3 |
+
optimizer: AdamW
|
| 4 |
+
encoder_learning_rate: 1.0e-05
|
| 5 |
+
learning_rate: 3.1e-05
|
| 6 |
+
layerwise_decay: 0.95
|
| 7 |
+
encoder_model: RoBERTa
|
| 8 |
+
pretrained_model: roberta-large
|
| 9 |
+
pool: avg
|
| 10 |
+
layer: mix
|
| 11 |
+
dropout: 0.15
|
| 12 |
+
batch_size: 4
|
| 13 |
+
class_identifier: referenceless_regression_metric
|
| 14 |
+
train_data: data/jan_27_zhen_num_0.5_del_None_mask_None_xlm_mbart.csv
|
| 15 |
+
validation_data: data/reference_less_zhen-comet.csv
|
| 16 |
+
hidden_sizes:
|
| 17 |
+
- 2048
|
| 18 |
+
- 1024
|
| 19 |
+
activations: Tanh
|
| 20 |
+
final_activation: null
|
| 21 |
+
load_weights_from_checkpoint: null
|