arslanarjumand
commited on
Commit
•
f35f35f
1
Parent(s):
1fe96c8
arslanarjumand/wav2vec-reptiles
Browse files- README.md +16 -23
- config.json +4 -4
- model.safetensors +1 -1
- training_args.bin +2 -2
README.md
CHANGED
@@ -15,11 +15,11 @@ should probably proofread and complete it, then remove this comment. -->
|
|
15 |
|
16 |
This model is a fine-tuned version of [arslanarjumand/wav2vec-reptiles](https://huggingface.co/arslanarjumand/wav2vec-reptiles) on the None dataset.
|
17 |
It achieves the following results on the evaluation set:
|
18 |
-
- Loss:
|
19 |
-
- Pcc Accuracy: 0.
|
20 |
-
- Pcc Fluency: 0.
|
21 |
-
- Pcc Total Score: 0.
|
22 |
-
- Pcc Content: 0.
|
23 |
|
24 |
## Model description
|
25 |
|
@@ -38,15 +38,15 @@ More information needed
|
|
38 |
### Training hyperparameters
|
39 |
|
40 |
The following hyperparameters were used during training:
|
41 |
-
- learning_rate: 5e-05
|
42 |
- train_batch_size: 4
|
43 |
- eval_batch_size: 6
|
44 |
- seed: 42
|
45 |
-
- gradient_accumulation_steps:
|
46 |
-
- total_train_batch_size:
|
47 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
48 |
- lr_scheduler_type: cosine
|
49 |
-
- lr_scheduler_warmup_ratio: 0.
|
50 |
- num_epochs: 15
|
51 |
- mixed_precision_training: Native AMP
|
52 |
|
@@ -54,20 +54,13 @@ The following hyperparameters were used during training:
|
|
54 |
|
55 |
| Training Loss | Epoch | Step | Validation Loss | Pcc Accuracy | Pcc Fluency | Pcc Total Score | Pcc Content |
|
56 |
|:-------------:|:-----:|:----:|:---------------:|:------------:|:-----------:|:---------------:|:-----------:|
|
57 |
-
|
|
58 |
-
|
|
59 |
-
|
|
60 |
-
|
|
61 |
-
|
|
62 |
-
|
|
63 |
-
|
|
64 |
-
| 512.1421 | 8.54 | 4000 | 355.2744 | 0.3539 | 0.3748 | 0.3857 | 0.3589 |
|
65 |
-
| 296.5867 | 9.61 | 4500 | 351.7932 | 0.3591 | 0.3816 | 0.3921 | 0.3656 |
|
66 |
-
| 316.3773 | 10.67 | 5000 | 350.8681 | 0.3622 | 0.3856 | 0.3960 | 0.3696 |
|
67 |
-
| 247.4901 | 11.74 | 5500 | 350.1711 | 0.3647 | 0.3893 | 0.3993 | 0.3731 |
|
68 |
-
| 262.0258 | 12.81 | 6000 | 348.5538 | 0.3658 | 0.3908 | 0.4007 | 0.3744 |
|
69 |
-
| 705.4405 | 13.87 | 6500 | 348.5071 | 0.3663 | 0.3917 | 0.4016 | 0.3754 |
|
70 |
-
| 264.0478 | 14.94 | 7000 | 348.5887 | 0.3663 | 0.3919 | 0.4017 | 0.3756 |
|
71 |
|
72 |
|
73 |
### Framework versions
|
|
|
15 |
|
16 |
This model is a fine-tuned version of [arslanarjumand/wav2vec-reptiles](https://huggingface.co/arslanarjumand/wav2vec-reptiles) on the None dataset.
|
17 |
It achieves the following results on the evaluation set:
|
18 |
+
- Loss: 180.5618
|
19 |
+
- Pcc Accuracy: 0.7344
|
20 |
+
- Pcc Fluency: 0.7572
|
21 |
+
- Pcc Total Score: 0.7949
|
22 |
+
- Pcc Content: 0.7727
|
23 |
|
24 |
## Model description
|
25 |
|
|
|
38 |
### Training hyperparameters
|
39 |
|
40 |
The following hyperparameters were used during training:
|
41 |
+
- learning_rate: 2.5e-05
|
42 |
- train_batch_size: 4
|
43 |
- eval_batch_size: 6
|
44 |
- seed: 42
|
45 |
+
- gradient_accumulation_steps: 4
|
46 |
+
- total_train_batch_size: 16
|
47 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
48 |
- lr_scheduler_type: cosine
|
49 |
+
- lr_scheduler_warmup_ratio: 0.5
|
50 |
- num_epochs: 15
|
51 |
- mixed_precision_training: Native AMP
|
52 |
|
|
|
54 |
|
55 |
| Training Loss | Epoch | Step | Validation Loss | Pcc Accuracy | Pcc Fluency | Pcc Total Score | Pcc Content |
|
56 |
|:-------------:|:-----:|:----:|:---------------:|:------------:|:-----------:|:---------------:|:-----------:|
|
57 |
+
| 323.2938 | 2.13 | 500 | 333.4772 | 0.4645 | 0.5166 | 0.5181 | 0.4915 |
|
58 |
+
| 274.2192 | 4.27 | 1000 | 259.5493 | 0.5725 | 0.6371 | 0.6430 | 0.6182 |
|
59 |
+
| 287.9362 | 6.4 | 1500 | 291.9187 | 0.6475 | 0.6895 | 0.7121 | 0.6902 |
|
60 |
+
| 273.6328 | 8.54 | 2000 | 229.1164 | 0.6884 | 0.7243 | 0.7522 | 0.7285 |
|
61 |
+
| 211.4504 | 10.67 | 2500 | 223.4485 | 0.7087 | 0.7420 | 0.7727 | 0.7499 |
|
62 |
+
| 162.7622 | 12.81 | 3000 | 180.6950 | 0.7302 | 0.7557 | 0.7918 | 0.7695 |
|
63 |
+
| 194.6916 | 14.94 | 3500 | 180.5618 | 0.7344 | 0.7572 | 0.7949 | 0.7727 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
|
66 |
### Framework versions
|
config.json
CHANGED
@@ -45,11 +45,11 @@
|
|
45 |
"layerdrop": 0.0005,
|
46 |
"left_max_position_embeddings": 64,
|
47 |
"mask_feature_length": 5,
|
48 |
-
"mask_feature_min_masks":
|
49 |
-
"mask_feature_prob": 0.
|
50 |
"mask_time_length": 5,
|
51 |
-
"mask_time_min_masks":
|
52 |
-
"mask_time_prob": 0.
|
53 |
"max_source_positions": 5000,
|
54 |
"model_type": "wav2vec2-bert",
|
55 |
"num_adapter_layers": 1,
|
|
|
45 |
"layerdrop": 0.0005,
|
46 |
"left_max_position_embeddings": 64,
|
47 |
"mask_feature_length": 5,
|
48 |
+
"mask_feature_min_masks": 5,
|
49 |
+
"mask_feature_prob": 0.0575,
|
50 |
"mask_time_length": 5,
|
51 |
+
"mask_time_min_masks": 5,
|
52 |
+
"mask_time_prob": 0.0585,
|
53 |
"max_source_positions": 5000,
|
54 |
"model_type": "wav2vec2-bert",
|
55 |
"num_adapter_layers": 1,
|
model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 2325236000
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3b0b673eb880d4c8d4ce1a725874267182d7bc3b1ff32d8b5061035cbe10c10a
|
3 |
size 2325236000
|
training_args.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ad61c98ac9e74083e7bf784e4b8953d284c8a3cf81d10f9c5fd2dfeec8b834da
|
3 |
+
size 4664
|