wav2vec2-large-xls-r-300m-sinhala-aug-data-with-original-split-part3

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0246
  • Wer: 0.0367

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: 0.0003
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 21

Training results

Training Loss Epoch Step Validation Loss Wer
5.8396 0.27 400 0.9752 0.8374
0.6967 0.54 800 0.3812 0.5935
0.4806 0.81 1200 0.3368 0.4757
0.3996 1.09 1600 0.1994 0.3143
0.3497 1.36 2000 0.1684 0.2564
0.3372 1.63 2400 0.1586 0.2419
0.312 1.9 2800 0.1413 0.2227
0.2797 2.17 3200 0.1466 0.2273
0.2554 2.44 3600 0.1543 0.2366
0.2613 2.71 4000 0.1450 0.2326
0.2399 2.99 4400 0.1238 0.2030
0.2125 3.26 4800 0.0989 0.1610
0.2144 3.53 5200 0.0984 0.1612
0.212 3.8 5600 0.0876 0.1507
0.1964 4.07 6000 0.1017 0.1753
0.1814 4.34 6400 0.0967 0.1654
0.1772 4.61 6800 0.0956 0.1631
0.1748 4.88 7200 0.0870 0.1483
0.1706 5.16 7600 0.0771 0.1306
0.1545 5.43 8000 0.0653 0.1199
0.1627 5.7 8400 0.0600 0.1103
0.1541 5.97 8800 0.0589 0.1068
0.1382 6.24 9200 0.0710 0.1231
0.1397 6.51 9600 0.0651 0.1248
0.1345 6.78 10000 0.0670 0.1194
0.1281 7.06 10400 0.0541 0.1006
0.1315 7.33 10800 0.0559 0.1062
0.1234 7.6 11200 0.0528 0.0970
0.1248 7.87 11600 0.0448 0.0865
0.115 8.14 12000 0.0546 0.0994
0.1143 8.41 12400 0.0595 0.1086
0.1169 8.68 12800 0.0485 0.0874
0.1165 8.96 13200 0.0524 0.0977
0.1035 9.23 13600 0.0445 0.0837
0.1017 9.5 14000 0.0413 0.0792
0.109 9.77 14400 0.0420 0.0833
0.1018 10.04 14800 0.0454 0.0823
0.0929 10.31 15200 0.0429 0.0786
0.0956 10.58 15600 0.0403 0.0772
0.0986 10.85 16000 0.0468 0.0906
0.0941 11.13 16400 0.0362 0.0694
0.0845 11.4 16800 0.0387 0.0702
0.0955 11.67 17200 0.0351 0.0627
0.089 11.94 17600 0.0361 0.0675
0.0806 12.21 18000 0.0381 0.0685
0.0803 12.48 18400 0.0370 0.0675
0.0839 12.75 18800 0.0333 0.0619
0.0834 13.03 19200 0.0334 0.0577
0.0779 13.3 19600 0.0358 0.0621
0.0773 13.57 20000 0.0330 0.0565
0.0717 13.84 20400 0.0350 0.0625
0.0737 14.11 20800 0.0355 0.0603
0.075 14.38 21200 0.0361 0.0626
0.0715 14.65 21600 0.0314 0.0575
0.0722 14.93 22000 0.0310 0.0575
0.0666 15.2 22400 0.0314 0.0559
0.0672 15.47 22800 0.0307 0.0535
0.0664 15.74 23200 0.0315 0.0552
0.0678 16.01 23600 0.0312 0.0548
0.0616 16.28 24000 0.0315 0.0527
0.0644 16.55 24400 0.0269 0.0481
0.062 16.82 24800 0.0308 0.0513
0.0584 17.1 25200 0.0294 0.0502
0.0563 17.37 25600 0.0294 0.0492
0.0547 17.64 26000 0.0281 0.0452
0.056 17.91 26400 0.0279 0.0451
0.0572 18.18 26800 0.0293 0.0460
0.0544 18.45 27200 0.0283 0.0464
0.052 18.72 27600 0.0274 0.0438
0.0533 19.0 28000 0.0264 0.0413
0.046 19.27 28400 0.0276 0.0412
0.0498 19.54 28800 0.0282 0.0419
0.0454 19.81 29200 0.0279 0.0417
0.0483 20.08 29600 0.0260 0.0396
0.0447 20.35 30000 0.0267 0.0418
0.0424 20.62 30400 0.0249 0.0373
0.0409 20.9 30800 0.0246 0.0367

Framework versions

  • Transformers 4.39.3
  • Pytorch 2.1.2
  • Datasets 2.18.0
  • Tokenizers 0.15.2
Downloads last month
11
Safetensors
Model size
316M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for SpideyDLK/wav2vec2-large-xls-r-300m-sinhala-aug-data-with-original-split-part3

Finetuned
(524)
this model