End of training

README.md

@@ -20,9 +20,9 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [google/madlad400-3b-mt](https://huggingface.co/google/madlad400-3b-mt) on the generator dataset.
 It achieves the following results on the evaluation set:
-- Loss: 5.2698
-- Bleu: 0.217
-- Gen Len: 9.6092
+- Loss: 4.4321
+- Bleu: 3.9461
+- Gen Len: 7.2243
 
 ## Model description
 
@@ -43,20 +43,19 @@ More information needed
 The following hyperparameters were used during training:
 - learning_rate: 0.0003
 - train_batch_size: 18
-- eval_batch_size: 18
+- eval_batch_size: 8
 - seed: 42
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: linear
-- num_epochs: 4
+- num_epochs: 3
 
 ### Training results
 
 | Training Loss | Epoch | Step  | Validation Loss | Bleu   | Gen Len |
 |:-------------:|:-----:|:-----:|:---------------:|:------:|:-------:|
-| 3.3025        | 1.0   | 2528  | 5.2043          | 0.2814 | 8.9524  |
-| 3.1868        | 2.0   | 5056  | 5.2746          | 0.1393 | 8.8278  |
-| 3.0907        | 3.0   | 7584  | 5.1989          | 0.2607 | 9.6087  |
-| 3.0409        | 4.0   | 10112 | 5.2698          | 0.217  | 9.6092  |
+| 2.1442        | 1.0   | 4997  | 4.4262          | 3.0781 | 7.4202  |
+| 2.0066        | 2.0   | 9994  | 4.4181          | 3.8977 | 7.2596  |
+| 1.948         | 3.0   | 14991 | 4.4321          | 3.9461 | 7.2243  |
 
 
 ### Framework versions