End of training

README.md

@@ -20,7 +20,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Common Voice 16.1 dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.9134
+- Loss: 0.8764
 
 ## Model description
 
@@ -48,15 +48,16 @@ The following hyperparameters were used during training:
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: linear
 - lr_scheduler_warmup_ratio: 0.1
-- num_epochs: 2
+- num_epochs: 4
 - mixed_precision_training: Native AMP
 
 ### Training results
 
-| Training Loss | Epoch | Step | Validation Loss |
-|:-------------:|:-----:|:----:|:---------------:|
-| 0.9769        | 1.0   | 2500 | 1.0105          |
-| 0.8011        | 2.0   | 5000 | 0.9134          |
+| Training Loss | Epoch | Step  | Validation Loss |
+|:-------------:|:-----:|:-----:|:---------------:|
+| 0.9769        | 1.0   | 2500  | 1.0105          |
+| 0.6894        | 3.0   | 7500  | 0.8839          |
+| 0.5316        | 4.0   | 10000 | 0.8764          |
 
 
 ### Framework versions

runs/Aug23_14-04-24_cd7ec1a0062f/events.out.tfevents.1724421865.cd7ec1a0062f.26.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15163e2c3357cfff8c33025a19bab748e6b94714c922e6a7bedd280a7d1f35ac
-size 47915
+oid sha256:9dcecead21f3ab6e69fba87d9d1407485e32c619add62df0380586f55f9fcc84
+size 48540