xls-r-uyghur-cv11

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_11_0 - UG dataset. It achieves the following results on the evaluation set:

• Loss: 0.2191
• Wer: 0.3257

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: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2000
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.3958	4.36	500	3.3137	1.0
3.032	8.71	1000	2.8586	0.9993
1.3977	13.07	1500	0.4786	0.6375
1.2751	17.43	2000	0.3816	0.5393
1.2113	21.79	2500	0.3451	0.5099
1.156	26.14	3000	0.3245	0.4919
1.1226	30.5	3500	0.2992	0.4441
1.0913	34.86	4000	0.2831	0.4315
1.0615	39.22	4500	0.2808	0.4340
1.0455	43.57	5000	0.2713	0.4088
1.0228	47.93	5500	0.2622	0.3960
0.9936	52.29	6000	0.2525	0.3796
0.968	56.64	6500	0.2506	0.3798
0.9704	61.0	7000	0.2481	0.3735
0.9552	65.36	7500	0.2394	0.3643
0.9417	69.72	8000	0.2350	0.3537
0.9215	74.07	8500	0.2326	0.3507
0.9097	78.43	9000	0.2277	0.3487
0.9003	82.79	9500	0.2230	0.3362
0.8857	87.15	10000	0.2246	0.3362
0.882	91.5	10500	0.2236	0.3315
0.8719	95.86	11000	0.2203	0.3271

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0