xlsr-am

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.8958
• Wer: 0.7517
• Cer: 0.2979

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: 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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
10.3463	3.0303	100	10.0110	1.0	1.0
4.3623	6.0606	200	4.3644	1.0	1.0
4.0873	9.0909	300	4.1130	1.0	0.9883
4.0428	12.1212	400	4.0742	1.0	0.9767
3.7646	15.1515	500	3.7872	1.0	0.9661
1.0192	18.1818	600	1.8339	0.9417	0.4527
0.4959	21.2121	700	1.6366	0.8650	0.3749
0.2754	24.2424	800	1.7326	0.8551	0.3833
0.2304	27.2727	900	1.8165	0.8699	0.3758
0.1876	30.3030	1000	1.8443	0.8260	0.3691
0.1889	33.3333	1100	1.7996	0.8248	0.3666
0.1787	36.3636	1200	1.8101	0.8067	0.3425
0.1099	39.3939	1300	1.8464	0.8256	0.3464
0.164	42.4242	1400	1.8417	0.7977	0.3337
0.1236	45.4545	1500	1.9786	0.8047	0.3414
0.1262	48.4848	1600	1.8451	0.7989	0.3344
0.1075	51.5152	1700	1.8909	0.7944	0.3369
0.1251	54.5455	1800	1.8848	0.8006	0.3252
0.0919	57.5758	1900	1.9425	0.7833	0.3230
0.0648	60.6061	2000	1.9384	0.7809	0.3262
0.0677	63.6364	2100	1.9723	0.7985	0.3338
0.0725	66.6667	2200	1.9917	0.7899	0.3292
0.0759	69.6970	2300	1.9070	0.7928	0.3210
0.0837	72.7273	2400	1.8595	0.7673	0.3103
0.0535	75.7576	2500	1.8503	0.7600	0.3086
0.0477	78.7879	2600	1.8926	0.7600	0.3109
0.0664	81.8182	2700	1.8792	0.7657	0.3100
0.0671	84.8485	2800	1.8895	0.7522	0.3038
0.0555	87.8788	2900	1.8770	0.7558	0.3055
0.027	90.9091	3000	1.9048	0.7583	0.3027
0.052	93.9394	3100	1.9058	0.7542	0.3015
0.0301	96.9697	3200	1.9015	0.7534	0.2996
0.027	100.0	3300	1.8958	0.7517	0.2979

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1