wav2vec2-xls-r-300m-dv

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

  • Loss: 0.2206
  • Wer: 0.2451

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: 0.0001
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 50
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
5.9623 0.66 400 3.3010 1.0
3.2238 1.33 800 2.8950 1.0
1.1988 1.99 1200 0.5277 0.6681
0.6084 2.65 1600 0.4113 0.5831
0.4973 3.32 2000 0.3538 0.5333
0.4476 3.98 2400 0.3201 0.5081
0.3999 4.64 2800 0.2917 0.4759
0.3779 5.31 3200 0.2788 0.4672
0.3457 5.97 3600 0.2667 0.4557
0.3222 6.63 4000 0.2549 0.4452
0.3129 7.3 4400 0.2491 0.4266
0.2927 7.96 4800 0.2488 0.4246
0.2786 8.62 5200 0.2429 0.4145
0.2756 9.29 5600 0.2453 0.4150
0.258 9.95 6000 0.2282 0.4109
0.251 10.61 6400 0.2307 0.4012
0.2397 11.28 6800 0.2275 0.4
0.2312 11.94 7200 0.2244 0.3889
0.2323 12.6 7600 0.2247 0.3983
0.216 13.27 8000 0.2301 0.3863
0.2169 13.93 8400 0.2224 0.3782
0.2089 14.59 8800 0.2276 0.3771
0.2042 15.26 9200 0.2286 0.3784
0.1953 15.92 9600 0.2235 0.3822
0.1876 16.58 10000 0.2267 0.3674
0.186 17.25 10400 0.2295 0.3676
0.1847 17.91 10800 0.2244 0.3608
0.178 18.57 11200 0.2229 0.3526
0.1751 19.24 11600 0.2219 0.3483
0.17 19.9 12000 0.2241 0.3503
0.1641 20.56 12400 0.2187 0.3403
0.1629 21.23 12800 0.2135 0.3433
0.1568 21.89 13200 0.2117 0.3358
0.1585 22.55 13600 0.2151 0.3332
0.1512 23.22 14000 0.2097 0.3344
0.1427 23.88 14400 0.2119 0.3255
0.1458 24.54 14800 0.2209 0.3213
0.1413 25.21 15200 0.2228 0.3202
0.1363 25.87 15600 0.2071 0.3207
0.1302 26.53 16000 0.2094 0.3138
0.1283 27.2 16400 0.2193 0.3132
0.1278 27.86 16800 0.2197 0.3103
0.1271 28.52 17200 0.2133 0.3009
0.1243 29.19 17600 0.2202 0.3026
0.1182 29.85 18000 0.2092 0.3046
0.1171 30.51 18400 0.2142 0.2947
0.1156 31.18 18800 0.2219 0.2926
0.1129 31.84 19200 0.2194 0.2848
0.1099 32.5 19600 0.2218 0.2869
0.1045 33.17 20000 0.2183 0.2803
0.1057 33.83 20400 0.2242 0.2896
0.1056 34.49 20800 0.2189 0.2838
0.1039 35.16 21200 0.2256 0.2819
0.1007 35.82 21600 0.2196 0.2743
0.1012 36.48 22000 0.2218 0.2752
0.098 37.15 22400 0.2181 0.2721
0.0963 37.81 22800 0.2162 0.2691
0.0943 38.47 23200 0.2148 0.2686
0.0959 39.14 23600 0.2194 0.2658
0.0904 39.8 24000 0.2170 0.2641
0.0898 40.46 24400 0.2129 0.2585
0.0886 41.13 24800 0.2199 0.2606
0.088 41.79 25200 0.2155 0.2595
0.0863 42.45 25600 0.2169 0.2564
0.0876 43.12 26000 0.2178 0.2529
0.0827 43.78 26400 0.2171 0.2559
0.087 44.44 26800 0.2192 0.2530
0.0818 45.11 27200 0.2180 0.2496
0.0811 45.77 27600 0.2207 0.2502
0.0828 46.43 28000 0.2186 0.2502
0.0796 47.1 28400 0.2203 0.2468
0.0804 47.76 28800 0.2201 0.2453
0.0791 48.42 29200 0.2204 0.2477
0.0777 49.09 29600 0.2197 0.2466
0.0775 49.75 30000 0.2206 0.2451

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu102
  • Datasets 1.18.3
  • Tokenizers 0.11.0
Downloads last month
14
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.

Dataset used to train shahukareem/wav2vec2-xls-r-300m-dv

Evaluation results