<!-- 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. -->
xtreme_s_xlsr_300m_fleurs_asr
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:
- Cer: 0.3330
- Loss: 1.2864
- Wer: 0.8344
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: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- total_train_batch_size: 32
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2000
- num_epochs: 5.0
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Cer | Validation Loss | Wer |
---|---|---|---|---|---|
4.677 | 0.13 | 1000 | 1.0 | 3.2323 | 1.0 |
4.1512 | 0.26 | 2000 | 0.5098 | 1.7858 | 0.9869 |
1.119 | 0.39 | 3000 | 0.4412 | 1.6628 | 0.9063 |
0.8573 | 0.52 | 4000 | 0.3588 | 1.3440 | 0.9016 |
1.0232 | 0.65 | 5000 | 0.3690 | 1.3004 | 0.8775 |
0.6328 | 0.78 | 6000 | 0.3354 | 1.2219 | 0.8331 |
0.6636 | 0.91 | 7000 | 0.3604 | 1.2839 | 0.8637 |
0.6536 | 1.04 | 8000 | 0.3420 | 1.2481 | 0.8504 |
0.5002 | 1.17 | 9000 | 0.3518 | 1.2514 | 0.8403 |
0.4785 | 1.3 | 10000 | 0.3399 | 1.2409 | 0.8570 |
0.517 | 1.43 | 11000 | 0.3599 | 1.3058 | 0.8654 |
0.506 | 1.56 | 12000 | 0.3484 | 1.2350 | 0.8441 |
0.4013 | 1.69 | 13000 | 0.3327 | 1.1982 | 0.8246 |
0.3521 | 1.82 | 14000 | 0.3270 | 1.1653 | 0.8265 |
0.4265 | 1.95 | 15000 | 0.3562 | 1.2647 | 0.8564 |
0.3949 | 2.08 | 16000 | 0.3490 | 1.2988 | 0.8480 |
0.3059 | 2.21 | 17000 | 0.3327 | 1.2332 | 0.8323 |
0.3618 | 2.34 | 18000 | 0.3480 | 1.2394 | 0.8517 |
0.2567 | 2.47 | 19000 | 0.3365 | 1.2294 | 0.8394 |
0.3501 | 2.6 | 20000 | 0.3271 | 1.1853 | 0.8250 |
0.2766 | 2.73 | 21000 | 0.3425 | 1.2339 | 0.8443 |
0.3396 | 2.86 | 22000 | 0.3501 | 1.2768 | 0.8669 |
0.3566 | 2.99 | 23000 | 0.3477 | 1.2648 | 0.8710 |
0.3166 | 3.12 | 24000 | 0.3550 | 1.3773 | 0.8641 |
0.2388 | 3.25 | 25000 | 0.3301 | 1.2374 | 0.8316 |
0.2057 | 3.38 | 26000 | 0.3429 | 1.2846 | 0.8560 |
0.2264 | 3.51 | 27000 | 0.3469 | 1.2676 | 0.8542 |
0.1998 | 3.64 | 28000 | 0.3531 | 1.3365 | 0.8655 |
0.2701 | 3.77 | 29000 | 0.3518 | 1.3124 | 0.8711 |
0.18 | 3.9 | 30000 | 0.3498 | 1.3095 | 0.8648 |
0.1337 | 4.03 | 31000 | 0.3397 | 1.2941 | 0.8452 |
0.162 | 4.16 | 32000 | 0.3320 | 1.2942 | 0.8295 |
0.2776 | 4.29 | 33000 | 0.3275 | 1.2690 | 0.8276 |
0.1634 | 4.42 | 34000 | 0.3307 | 1.3145 | 0.8331 |
0.2172 | 4.54 | 35000 | 0.3334 | 1.3031 | 0.8435 |
0.1305 | 4.67 | 36000 | 0.3303 | 1.2768 | 0.8321 |
0.1436 | 4.8 | 37000 | 0.3353 | 1.2968 | 0.8416 |
0.134 | 4.93 | 38000 | 0.3330 | 1.2864 | 0.8344 |
Framework versions
- Transformers 4.18.0.dev0
- Pytorch 1.10.1+cu111
- Datasets 1.18.4.dev0
- Tokenizers 0.11.6