Jan van Doorn
commited on
Commit
•
f529912
1
Parent(s):
0814c0b
update model card README.md
Browse files
README.md
CHANGED
@@ -14,10 +14,10 @@ should probably proofread and complete it, then remove this comment. -->
|
|
14 |
|
15 |
# whisper-large-v2-atcosim
|
16 |
|
17 |
-
This model is a fine-tuned version of [openai/whisper-large-v2](https://huggingface.co/openai/whisper-large-v2) on the
|
18 |
It achieves the following results on the evaluation set:
|
19 |
-
- Loss: 0.
|
20 |
-
- Wer:
|
21 |
|
22 |
## Model description
|
23 |
|
@@ -41,65 +41,30 @@ The following hyperparameters were used during training:
|
|
41 |
- eval_batch_size: 8
|
42 |
- seed: 42
|
43 |
- distributed_type: multi-GPU
|
|
|
|
|
|
|
44 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
45 |
- lr_scheduler_type: linear
|
46 |
-
- lr_scheduler_warmup_steps:
|
47 |
-
- training_steps:
|
48 |
|
49 |
### Training results
|
50 |
|
51 |
-
| Training Loss | Epoch
|
52 |
-
|
53 |
-
| 0.
|
54 |
-
| 0.
|
55 |
-
| 0.
|
56 |
-
| 0.
|
57 |
-
| 0.
|
58 |
-
| 0.
|
59 |
-
| 0.
|
60 |
-
| 0.
|
61 |
-
| 0.
|
62 |
-
| 0.
|
63 |
-
| 0.
|
64 |
-
| 0.
|
65 |
-
| 0.004 | 27.2 | 13000 | 0.0442 | 2.8385 |
|
66 |
-
| 0.0018 | 29.29 | 14000 | 0.0444 | 2.5282 |
|
67 |
-
| 0.0011 | 31.38 | 15000 | 0.0467 | 4.0980 |
|
68 |
-
| 0.0002 | 33.47 | 16000 | 0.0469 | 3.9128 |
|
69 |
-
| 0.003 | 35.56 | 17000 | 0.0454 | 4.7462 |
|
70 |
-
| 0.0001 | 37.66 | 18000 | 0.0459 | 3.1950 |
|
71 |
-
| 0.0006 | 39.75 | 19000 | 0.0451 | 2.6579 |
|
72 |
-
| 0.0014 | 41.84 | 20000 | 0.0464 | 1.6855 |
|
73 |
-
| 0.0 | 43.93 | 21000 | 0.0487 | 2.3106 |
|
74 |
-
| 0.0005 | 46.03 | 22000 | 0.0535 | 7.3717 |
|
75 |
-
| 0.0001 | 48.12 | 23000 | 0.0482 | 6.9411 |
|
76 |
-
| 0.0002 | 50.21 | 24000 | 0.0484 | 13.0580 |
|
77 |
-
| 0.0001 | 52.3 | 25000 | 0.0481 | 18.0219 |
|
78 |
-
| 0.0 | 54.39 | 26000 | 0.0523 | 14.7342 |
|
79 |
-
| 0.0 | 56.49 | 27000 | 0.0552 | 11.1132 |
|
80 |
-
| 0.0004 | 58.58 | 28000 | 0.0521 | 2.5190 |
|
81 |
-
| 0.0001 | 60.67 | 29000 | 0.0490 | 4.4036 |
|
82 |
-
| 0.0 | 62.76 | 30000 | 0.0497 | 2.8246 |
|
83 |
-
| 0.0 | 64.85 | 31000 | 0.0513 | 2.8755 |
|
84 |
-
| 0.0 | 66.95 | 32000 | 0.0526 | 2.9172 |
|
85 |
-
| 0.0 | 69.04 | 33000 | 0.0539 | 3.0098 |
|
86 |
-
| 0.0 | 71.13 | 34000 | 0.0552 | 3.0144 |
|
87 |
-
| 0.0 | 73.22 | 35000 | 0.0566 | 3.1209 |
|
88 |
-
| 0.0 | 75.31 | 36000 | 0.0580 | 3.2321 |
|
89 |
-
| 0.0 | 77.41 | 37000 | 0.0594 | 3.4729 |
|
90 |
-
| 0.0 | 79.5 | 38000 | 0.0607 | 3.6164 |
|
91 |
-
| 0.0 | 81.59 | 39000 | 0.0620 | 3.9035 |
|
92 |
-
| 0.0 | 83.68 | 40000 | 0.0632 | 4.0656 |
|
93 |
-
| 0.0 | 85.77 | 41000 | 0.0642 | 4.3202 |
|
94 |
-
| 0.0 | 87.87 | 42000 | 0.0651 | 4.4453 |
|
95 |
-
| 0.0 | 89.96 | 43000 | 0.0659 | 4.9361 |
|
96 |
-
| 0.0 | 92.05 | 44000 | 0.0664 | 5.2186 |
|
97 |
-
| 0.0 | 94.14 | 45000 | 0.0670 | 5.6029 |
|
98 |
-
| 0.0 | 96.23 | 46000 | 0.0673 | 5.7835 |
|
99 |
-
| 0.0 | 98.33 | 47000 | 0.0676 | 6.0520 |
|
100 |
-
| 0.0 | 100.42 | 48000 | 0.0678 | 6.1122 |
|
101 |
-
| 0.0 | 102.51 | 49000 | 0.0679 | 6.2141 |
|
102 |
-
| 0.0 | 104.6 | 50000 | 0.0679 | 6.2234 |
|
103 |
|
104 |
|
105 |
### Framework versions
|
|
|
14 |
|
15 |
# whisper-large-v2-atcosim
|
16 |
|
17 |
+
This model is a fine-tuned version of [openai/whisper-large-v2](https://huggingface.co/openai/whisper-large-v2) on the None dataset.
|
18 |
It achieves the following results on the evaluation set:
|
19 |
+
- Loss: 0.0552
|
20 |
+
- Wer: 9.9694
|
21 |
|
22 |
## Model description
|
23 |
|
|
|
41 |
- eval_batch_size: 8
|
42 |
- seed: 42
|
43 |
- distributed_type: multi-GPU
|
44 |
+
- num_devices: 4
|
45 |
+
- total_train_batch_size: 64
|
46 |
+
- total_eval_batch_size: 32
|
47 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
48 |
- lr_scheduler_type: linear
|
49 |
+
- lr_scheduler_warmup_steps: 250
|
50 |
+
- training_steps: 12500
|
51 |
|
52 |
### Training results
|
53 |
|
54 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
55 |
+
|:-------------:|:-----:|:-----:|:---------------:|:-------:|
|
56 |
+
| 0.0038 | 8.33 | 1000 | 0.0357 | 2.7829 |
|
57 |
+
| 0.001 | 16.67 | 2000 | 0.0384 | 2.0004 |
|
58 |
+
| 0.0015 | 25.0 | 3000 | 0.0373 | 31.7142 |
|
59 |
+
| 0.0001 | 33.33 | 4000 | 0.0437 | 2.3152 |
|
60 |
+
| 0.0019 | 41.67 | 5000 | 0.0446 | 7.2375 |
|
61 |
+
| 0.0 | 50.0 | 6000 | 0.0462 | 2.9033 |
|
62 |
+
| 0.0 | 58.33 | 7000 | 0.0490 | 4.3295 |
|
63 |
+
| 0.0 | 66.67 | 8000 | 0.0509 | 5.8668 |
|
64 |
+
| 0.0 | 75.0 | 9000 | 0.0524 | 7.5014 |
|
65 |
+
| 0.0 | 83.33 | 10000 | 0.0536 | 8.6405 |
|
66 |
+
| 0.0 | 91.67 | 11000 | 0.0546 | 9.5018 |
|
67 |
+
| 0.0 | 100.0 | 12000 | 0.0552 | 9.9694 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
|
70 |
### Framework versions
|