update model card README.md
Browse files
README.md
CHANGED
@@ -21,7 +21,7 @@ model-index:
|
|
21 |
metrics:
|
22 |
- name: Wer
|
23 |
type: wer
|
24 |
-
value: 0.
|
25 |
---
|
26 |
|
27 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
@@ -31,8 +31,8 @@ should probably proofread and complete it, then remove this comment. -->
|
|
31 |
|
32 |
This model is a fine-tuned version of [airesearch/wav2vec2-large-xlsr-53-th](https://huggingface.co/airesearch/wav2vec2-large-xlsr-53-th) on the common_voice dataset.
|
33 |
It achieves the following results on the evaluation set:
|
34 |
-
- Loss: 1.
|
35 |
-
- Wer: 0.
|
36 |
|
37 |
## Model description
|
38 |
|
@@ -58,44 +58,23 @@ The following hyperparameters were used during training:
|
|
58 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
59 |
- lr_scheduler_type: linear
|
60 |
- lr_scheduler_warmup_steps: 30
|
61 |
-
- num_epochs:
|
62 |
- mixed_precision_training: Native AMP
|
63 |
|
64 |
### Training results
|
65 |
|
66 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
67 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
68 |
-
| 5.
|
69 |
-
| 3.
|
70 |
-
| 2.
|
71 |
-
| 1.
|
72 |
-
| 0.
|
73 |
-
| 0.
|
74 |
-
| 0.
|
75 |
-
| 0.
|
76 |
-
| 0.
|
77 |
-
| 0.
|
78 |
-
| 0.4587 | 35.48 | 1100 | 1.3144 | 0.4722 |
|
79 |
-
| 0.4005 | 38.71 | 1200 | 1.3111 | 0.4686 |
|
80 |
-
| 0.3851 | 41.94 | 1300 | 1.3420 | 0.4786 |
|
81 |
-
| 0.3563 | 45.16 | 1400 | 1.3679 | 0.4743 |
|
82 |
-
| 0.3591 | 48.39 | 1500 | 1.4444 | 0.4643 |
|
83 |
-
| 0.325 | 51.61 | 1600 | 1.4076 | 0.4722 |
|
84 |
-
| 0.3409 | 54.84 | 1700 | 1.4586 | 0.4629 |
|
85 |
-
| 0.3019 | 58.06 | 1800 | 1.4579 | 0.4529 |
|
86 |
-
| 0.292 | 61.29 | 1900 | 1.4887 | 0.4522 |
|
87 |
-
| 0.2729 | 64.52 | 2000 | 1.4966 | 0.4608 |
|
88 |
-
| 0.2656 | 67.74 | 2100 | 1.5232 | 0.4593 |
|
89 |
-
| 0.2575 | 70.97 | 2200 | 1.4984 | 0.4508 |
|
90 |
-
| 0.2532 | 74.19 | 2300 | 1.5332 | 0.4544 |
|
91 |
-
| 0.2474 | 77.42 | 2400 | 1.5301 | 0.4529 |
|
92 |
-
| 0.2539 | 80.65 | 2500 | 1.5214 | 0.4601 |
|
93 |
-
| 0.2526 | 83.87 | 2600 | 1.5413 | 0.4572 |
|
94 |
-
| 0.2601 | 87.1 | 2700 | 1.5553 | 0.4608 |
|
95 |
-
| 0.2315 | 90.32 | 2800 | 1.5768 | 0.4515 |
|
96 |
-
| 0.2477 | 93.55 | 2900 | 1.5787 | 0.4650 |
|
97 |
-
| 0.2363 | 96.77 | 3000 | 1.5900 | 0.4565 |
|
98 |
-
| 0.242 | 100.0 | 3100 | 1.5867 | 0.4551 |
|
99 |
|
100 |
|
101 |
### Framework versions
|
|
|
21 |
metrics:
|
22 |
- name: Wer
|
23 |
type: wer
|
24 |
+
value: 0.4557774607703281
|
25 |
---
|
26 |
|
27 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
|
|
31 |
|
32 |
This model is a fine-tuned version of [airesearch/wav2vec2-large-xlsr-53-th](https://huggingface.co/airesearch/wav2vec2-large-xlsr-53-th) on the common_voice dataset.
|
33 |
It achieves the following results on the evaluation set:
|
34 |
+
- Loss: 1.0926
|
35 |
+
- Wer: 0.4558
|
36 |
|
37 |
## Model description
|
38 |
|
|
|
58 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
59 |
- lr_scheduler_type: linear
|
60 |
- lr_scheduler_warmup_steps: 30
|
61 |
+
- num_epochs: 35
|
62 |
- mixed_precision_training: Native AMP
|
63 |
|
64 |
### Training results
|
65 |
|
66 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
67 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
68 |
+
| 5.4829 | 3.23 | 100 | 3.4277 | 1.0 |
|
69 |
+
| 3.301 | 6.45 | 200 | 3.2260 | 1.0 |
|
70 |
+
| 2.3 | 9.68 | 300 | 1.2590 | 0.5292 |
|
71 |
+
| 1.1601 | 12.9 | 400 | 1.0578 | 0.5892 |
|
72 |
+
| 0.9381 | 16.13 | 500 | 1.0390 | 0.4957 |
|
73 |
+
| 0.8211 | 19.35 | 600 | 1.0411 | 0.4679 |
|
74 |
+
| 0.7227 | 22.58 | 700 | 1.0614 | 0.4558 |
|
75 |
+
| 0.6805 | 25.81 | 800 | 1.0750 | 0.4579 |
|
76 |
+
| 0.6389 | 29.03 | 900 | 1.0678 | 0.4629 |
|
77 |
+
| 0.6045 | 32.26 | 1000 | 1.0926 | 0.4558 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
|
79 |
|
80 |
### Framework versions
|