Update README.md

README.md

@@ -31,17 +31,16 @@ should probably proofread and complete it, then remove this comment. -->
 # xls-asr-vi-40h-1B
 
 This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the common voice 7.0 vi & private dataset.
-It achieves the following results on the evaluation set Common voice 7.0 vi test set (Without Language Model):
-- Cer: 19.94
-- Wer: 34.21
 
-## Model description
+### Benchmark WER result:
+| | [VIVOS](https://huggingface.co/datasets/vivos) | [COMMON VOICE 7.0 VI](https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0) |
+|---|---|---|
+|without LM| 25.92 | 34.21 |
 
-More information needed
-
-## Intended uses & limitations
-
-More information needed
+### Benchmark CER result:
+| | [VIVOS](https://huggingface.co/datasets/vivos) | [COMMON VOICE 7.0 VI](https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0) |
+|---|---|---|
+|without LM| 9.24 | 19.94 |
 
 ## Training and evaluation data
 
@@ -63,6 +62,13 @@ The following hyperparameters were used during training:
 - lr_scheduler_warmup_steps: 1500
 - num_epochs: 50.0
 - mixed_precision_training: Native AMP
+- attention_dropout: 0.2
+- activation_dropout: 0.1
+- warmup_steps: 1500
+- mask_time_prob: .15
+- mask_time_length: 10
+- mask_feature_prob: 0.25
+- mask_feature_length: 64
 
 ### Training results