<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
whisper-small-emergency
This model is a fine-tuned version of openai/whisper-small on the whisper-small-kor dataset. It achieves the following results on the evaluation set:
- Loss: 0.2212
- Wer: 21.7895
- Cer: 10.3463
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- training_steps: 10000
Training results
Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
---|---|---|---|---|---|
2.2586 | 0.02 | 100 | 2.0061 | 38.3870 | 19.3958 |
0.9821 | 0.05 | 200 | 0.8927 | 37.3486 | 16.9619 |
0.7244 | 0.07 | 300 | 0.6577 | 32.9353 | 15.2939 |
0.505 | 0.1 | 400 | 0.4395 | 33.8006 | 16.8397 |
0.4397 | 0.12 | 500 | 0.3908 | 29.2489 | 13.3393 |
0.3602 | 0.15 | 600 | 0.3710 | 27.7259 | 12.8741 |
0.4321 | 0.17 | 700 | 0.3558 | 27.7778 | 12.8224 |
0.3979 | 0.19 | 800 | 0.3504 | 27.5528 | 12.6063 |
0.2614 | 0.22 | 900 | 0.3434 | 28.2451 | 13.5601 |
0.3725 | 0.24 | 1000 | 0.3362 | 26.8086 | 12.8177 |
0.4098 | 0.27 | 1100 | 0.3329 | 26.8086 | 13.1466 |
0.3083 | 0.29 | 1200 | 0.3240 | 25.6663 | 12.0566 |
0.324 | 0.32 | 1300 | 0.3169 | 24.7490 | 11.3659 |
0.3437 | 0.34 | 1400 | 0.3090 | 24.2471 | 10.9383 |
0.3719 | 0.36 | 1500 | 0.3064 | 24.4548 | 11.2155 |
0.3563 | 0.39 | 1600 | 0.3013 | 24.0222 | 11.0887 |
0.3493 | 0.41 | 1700 | 0.3036 | 24.1606 | 11.1779 |
0.3132 | 0.44 | 1800 | 0.3011 | 24.0741 | 11.1685 |
0.3024 | 0.46 | 1900 | 0.2920 | 24.4202 | 11.2014 |
0.2982 | 0.49 | 2000 | 0.2873 | 22.9664 | 10.4262 |
0.3309 | 0.51 | 2100 | 0.2880 | 23.3991 | 10.8208 |
0.3209 | 0.53 | 2200 | 0.2811 | 21.9280 | 10.2288 |
0.2778 | 0.56 | 2300 | 0.2883 | 22.6895 | 10.5060 |
0.3391 | 0.58 | 2400 | 0.2796 | 21.9280 | 10.1818 |
0.3261 | 0.61 | 2500 | 0.2757 | 22.3607 | 10.1865 |
0.2711 | 0.63 | 2600 | 0.2746 | 22.9491 | 10.4356 |
0.2723 | 0.66 | 2700 | 0.2708 | 22.3088 | 10.5624 |
0.3152 | 0.68 | 2800 | 0.2681 | 21.8934 | 10.0127 |
0.248 | 0.7 | 2900 | 0.2679 | 22.2568 | 10.0644 |
0.2354 | 0.73 | 3000 | 0.2665 | 21.7203 | 9.8576 |
0.2828 | 0.75 | 3100 | 0.2628 | 21.4261 | 9.9422 |
0.2759 | 0.78 | 3200 | 0.2652 | 21.2703 | 9.8623 |
0.2904 | 0.8 | 3300 | 0.2606 | 21.2876 | 9.8388 |
0.2844 | 0.83 | 3400 | 0.2600 | 21.8761 | 10.0362 |
0.2815 | 0.85 | 3500 | 0.2554 | 20.9069 | 9.5992 |
0.2713 | 0.87 | 3600 | 0.2573 | 20.8550 | 9.5334 |
0.2748 | 0.9 | 3700 | 0.2566 | 21.5126 | 9.8811 |
0.2447 | 0.92 | 3800 | 0.2526 | 20.5088 | 9.3455 |
0.3255 | 0.95 | 3900 | 0.2517 | 20.3358 | 11.3048 |
0.2786 | 0.97 | 4000 | 0.2489 | 20.8030 | 9.5898 |
0.245 | 1.0 | 4100 | 0.2523 | 21.4607 | 9.7167 |
0.1655 | 1.02 | 4200 | 0.2470 | 20.4396 | 9.5287 |
0.1898 | 1.04 | 4300 | 0.2422 | 19.9550 | 9.0871 |
0.1394 | 1.07 | 4400 | 0.2429 | 20.0242 | 9.2750 |
0.1592 | 1.09 | 4500 | 0.2433 | 19.9896 | 9.0824 |
0.1542 | 1.12 | 4600 | 0.2428 | 20.2492 | 9.3126 |
0.1296 | 1.14 | 4700 | 0.2437 | 19.4531 | 8.9038 |
0.1477 | 1.17 | 4800 | 0.2432 | 19.7300 | 11.0605 |
0.1551 | 1.19 | 4900 | 0.2436 | 20.0762 | 11.3236 |
0.1581 | 1.21 | 5000 | 0.2435 | 19.7992 | 10.9994 |
0.2033 | 1.24 | 5100 | 0.2434 | 19.8339 | 9.1763 |
0.1444 | 1.26 | 5200 | 0.2399 | 19.8165 | 10.9806 |
0.1543 | 1.29 | 5300 | 0.2371 | 19.1762 | 10.8913 |
0.1735 | 1.31 | 5400 | 0.2350 | 19.4185 | 9.0166 |
0.1552 | 1.34 | 5500 | 0.2363 | 19.0897 | 8.8098 |
0.1495 | 1.36 | 5600 | 0.2332 | 19.1070 | 8.8145 |
0.1636 | 1.38 | 5700 | 0.2350 | 18.6051 | 10.5718 |
0.1827 | 1.41 | 5800 | 0.2333 | 18.4493 | 8.5091 |
0.1464 | 1.43 | 5900 | 0.2344 | 19.2454 | 8.8850 |
0.1999 | 1.46 | 6000 | 0.2325 | 23.1222 | 10.9900 |
0.1547 | 1.48 | 6100 | 0.2318 | 19.3839 | 8.8709 |
0.1296 | 1.51 | 6200 | 0.2339 | 19.3146 | 8.9085 |
0.1535 | 1.53 | 6300 | 0.2317 | 22.5684 | 10.8302 |
0.1467 | 1.55 | 6400 | 0.2310 | 19.1070 | 8.7958 |
0.1709 | 1.58 | 6500 | 0.2338 | 18.9685 | 8.7441 |
0.1359 | 1.6 | 6600 | 0.2295 | 19.0550 | 8.6548 |
0.1611 | 1.63 | 6700 | 0.2293 | 18.5877 | 8.5608 |
0.1232 | 1.65 | 6800 | 0.2309 | 19.4012 | 8.9273 |
0.1692 | 1.68 | 6900 | 0.2288 | 18.6224 | 8.8756 |
0.1544 | 1.7 | 7000 | 0.2265 | 18.3454 | 8.5467 |
0.1282 | 1.72 | 7100 | 0.2256 | 18.6570 | 8.6642 |
0.1414 | 1.75 | 7200 | 0.2258 | 22.1011 | 10.2993 |
0.157 | 1.77 | 7300 | 0.2259 | 18.8474 | 8.6501 |
0.1592 | 1.8 | 7400 | 0.2249 | 18.6570 | 8.5702 |
0.0998 | 1.82 | 7500 | 0.2246 | 18.8127 | 8.6125 |
0.1486 | 1.85 | 7600 | 0.2225 | 18.3281 | 8.3024 |
0.1336 | 1.87 | 7700 | 0.2221 | 18.5704 | 8.4387 |
0.1388 | 1.9 | 7800 | 0.2222 | 18.5531 | 8.5044 |
0.1341 | 1.92 | 7900 | 0.2212 | 22.0665 | 10.4215 |
0.1548 | 1.94 | 8000 | 0.2215 | 21.8588 | 10.3275 |
0.1276 | 1.97 | 8100 | 0.2182 | 21.8069 | 10.3040 |
0.1567 | 1.99 | 8200 | 0.2200 | 18.1031 | 8.3541 |
0.1054 | 2.02 | 8300 | 0.2201 | 21.5646 | 10.2335 |
0.0793 | 2.04 | 8400 | 0.2219 | 21.1838 | 10.1161 |
0.0944 | 2.07 | 8500 | 0.2225 | 21.5819 | 10.3510 |
0.0824 | 2.09 | 8600 | 0.2230 | 21.7203 | 10.2476 |
0.0863 | 2.11 | 8700 | 0.2222 | 21.6684 | 10.2241 |
0.1102 | 2.14 | 8800 | 0.2233 | 21.5819 | 10.3228 |
0.0852 | 2.16 | 8900 | 0.2226 | 21.8588 | 10.2946 |
0.0796 | 2.19 | 9000 | 0.2227 | 21.9626 | 10.3651 |
0.1023 | 2.21 | 9100 | 0.2223 | 21.7722 | 10.4309 |
0.08 | 2.24 | 9200 | 0.2216 | 21.4780 | 10.2664 |
0.0703 | 2.26 | 9300 | 0.2218 | 21.5992 | 10.2429 |
0.0923 | 2.28 | 9400 | 0.2212 | 21.4434 | 10.2006 |
0.0694 | 2.31 | 9500 | 0.2217 | 21.4780 | 10.2194 |
0.1033 | 2.33 | 9600 | 0.2216 | 21.5126 | 10.2382 |
0.0913 | 2.36 | 9700 | 0.2214 | 21.5299 | 10.2194 |
0.0882 | 2.38 | 9800 | 0.2212 | 21.7376 | 10.2758 |
0.0852 | 2.41 | 9900 | 0.2212 | 21.7203 | 10.3087 |
0.0862 | 2.43 | 10000 | 0.2212 | 21.7895 | 10.3463 |
Framework versions
- Transformers 4.33.2
- Pytorch 2.0.1+cu117
- Datasets 2.14.5
- Tokenizers 0.13.3