Commit
·
eb0fc7e
1
Parent(s):
0197697
update model card README.md
Browse files
README.md
CHANGED
|
@@ -16,8 +16,13 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 16 |
|
| 17 |
This model is a fine-tuned version of [facebook/wav2vec2-base](https://huggingface.co/facebook/wav2vec2-base) on the timit_asr dataset.
|
| 18 |
It achieves the following results on the evaluation set:
|
| 19 |
-
-
|
| 20 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
## Model description
|
| 23 |
|
|
@@ -43,39 +48,9 @@ The following hyperparameters were used during training:
|
|
| 43 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 44 |
- lr_scheduler_type: linear
|
| 45 |
- lr_scheduler_warmup_steps: 1000
|
| 46 |
-
- num_epochs:
|
| 47 |
- mixed_precision_training: Native AMP
|
| 48 |
|
| 49 |
-
### Training results
|
| 50 |
-
|
| 51 |
-
| Training Loss | Epoch | Step | Validation Loss | Cer |
|
| 52 |
-
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
| 53 |
-
| 6.1575 | 0.4 | 100 | 3.4447 | 0.9385 |
|
| 54 |
-
| 2.7081 | 0.8 | 200 | 1.7826 | 0.6131 |
|
| 55 |
-
| 1.0878 | 1.2 | 300 | 0.7552 | 0.2370 |
|
| 56 |
-
| 0.6164 | 1.61 | 400 | 0.5745 | 0.2017 |
|
| 57 |
-
| 0.5154 | 2.01 | 500 | 0.7112 | 0.2224 |
|
| 58 |
-
| 0.4486 | 2.41 | 600 | 0.4809 | 0.1668 |
|
| 59 |
-
| 0.4285 | 2.81 | 700 | 0.6632 | 0.1550 |
|
| 60 |
-
| 0.401 | 3.21 | 800 | 0.5330 | 0.1638 |
|
| 61 |
-
| 0.3969 | 3.61 | 900 | 0.3972 | 0.1349 |
|
| 62 |
-
| 0.3871 | 4.02 | 1000 | 0.3583 | 0.1386 |
|
| 63 |
-
| 0.3642 | 4.42 | 1100 | 0.3638 | 0.1319 |
|
| 64 |
-
| 0.3413 | 4.82 | 1200 | 0.3620 | 0.1298 |
|
| 65 |
-
| 0.3212 | 5.22 | 1300 | 0.3432 | 0.1326 |
|
| 66 |
-
| 0.3119 | 5.62 | 1400 | 0.3407 | 0.1271 |
|
| 67 |
-
| 0.2896 | 6.02 | 1500 | 0.3397 | 0.1252 |
|
| 68 |
-
| 0.2533 | 6.43 | 1600 | 0.3186 | 0.1249 |
|
| 69 |
-
| 0.2614 | 6.83 | 1700 | 0.3256 | 0.1221 |
|
| 70 |
-
| 0.2415 | 7.23 | 1800 | 0.3400 | 0.1211 |
|
| 71 |
-
| 0.2216 | 7.63 | 1900 | 0.3378 | 0.1203 |
|
| 72 |
-
| 0.2268 | 8.03 | 2000 | 0.3928 | 0.1227 |
|
| 73 |
-
| 0.1968 | 8.43 | 2100 | 0.3601 | 0.1199 |
|
| 74 |
-
| 0.1939 | 8.84 | 2200 | 0.3871 | 0.1207 |
|
| 75 |
-
| 0.1918 | 9.24 | 2300 | 0.3648 | 0.1198 |
|
| 76 |
-
| 0.1822 | 9.64 | 2400 | 0.3762 | 0.1183 |
|
| 77 |
-
|
| 78 |
-
|
| 79 |
### Framework versions
|
| 80 |
|
| 81 |
- Transformers 4.17.0
|
|
|
|
| 16 |
|
| 17 |
This model is a fine-tuned version of [facebook/wav2vec2-base](https://huggingface.co/facebook/wav2vec2-base) on the timit_asr dataset.
|
| 18 |
It achieves the following results on the evaluation set:
|
| 19 |
+
- eval_loss: 0.4126
|
| 20 |
+
- eval_cer: 0.1224
|
| 21 |
+
- eval_runtime: 17.9127
|
| 22 |
+
- eval_samples_per_second: 46.894
|
| 23 |
+
- eval_steps_per_second: 5.862
|
| 24 |
+
- epoch: 15.26
|
| 25 |
+
- step: 3800
|
| 26 |
|
| 27 |
## Model description
|
| 28 |
|
|
|
|
| 48 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 49 |
- lr_scheduler_type: linear
|
| 50 |
- lr_scheduler_warmup_steps: 1000
|
| 51 |
+
- num_epochs: 25
|
| 52 |
- mixed_precision_training: Native AMP
|
| 53 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
### Framework versions
|
| 55 |
|
| 56 |
- Transformers 4.17.0
|