Update model card with final test metrics

README.md

@@ -13,7 +13,7 @@ base_model: openai/whisper-small
 
 Fine-tuned `openai/whisper-small` for Santali IPA/Sanlish transliteration (speech-to-text).
 
-Last updated: 2026-08-29T23:56:14.166366Z — after epoch 20.0
+Last updated: 2026-08-29T23:56:15.284469Z — after epoch 20.0
 
 ## Dataset split (80/10/10)
 - train: 1547
@@ -23,10 +23,6 @@ Last updated: 2026-08-29T23:56:14.166366Z — after epoch 20.0
 ## Latest evaluation metrics (epoch 20.0)
 - **test_loss**: 0.4051657021045685
 - **test_wer**: 40.782828282828284
-- **test_runtime**: 95.9814
-- **test_samples_per_second**: 2.021
-- **test_steps_per_second**: 0.26
-- **epoch**: 20.0
 
 ## Training hyperparameters
 - per_device_train_batch_size: 8