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: 0.8792
+- Loss: 0.8441
 
 ## 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 |
-|:-------------:|:-----:|:----:|:---------------:|
-| 1.1071        | 1.0   | 2500 | 0.9696          |
-| 0.8146        | 2.0   | 5000 | 0.8792          |
+| Training Loss | Epoch | Step  | Validation Loss |
+|:-------------:|:-----:|:-----:|:---------------:|
+| 1.1071        | 1.0   | 2500  | 0.9696          |
+| 0.7374        | 3.0   | 7500  | 0.8514          |
+| 0.5902        | 4.0   | 10000 | 0.8441          |
 
 
 ### Framework versions

runs/Oct05_18-43-18_444f3fc62787/events.out.tfevents.1728153800.444f3fc62787.27.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c03e6d80e257228b2ba7358b54db681e30e89e0e52d02274344e5047cd048423
-size 47928
+oid sha256:427867eb7709bdfd46845807c7c58410029a8a6ccca5bed1cd02b0aff0f33854
+size 48553