xlsr-am-adap-ar

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: 1.6979
• Wer: 0.6968
• Cer: 0.2762

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
4.7315	3.0303	100	4.5550	1.0	1.0
4.1365	6.0606	200	4.1643	1.0	0.9883
3.2632	9.0909	300	3.2750	0.9996	0.8189
1.3193	12.1212	400	1.7260	0.9146	0.4165
0.7873	15.1515	500	1.5051	0.8658	0.3773
0.7762	18.1818	600	1.4205	0.8014	0.3242
0.4153	21.2121	700	1.4724	0.8006	0.3200
0.3188	24.2424	800	1.4552	0.7682	0.3034
0.3318	27.2727	900	1.4800	0.7690	0.3010
0.2333	30.3030	1000	1.5290	0.7452	0.2903
0.2029	33.3333	1100	1.5773	0.7661	0.3207
0.1871	36.3636	1200	1.5868	0.7628	0.3004
0.2034	39.3939	1300	1.6034	0.7403	0.2931
0.1979	42.4242	1400	1.5973	0.7542	0.3052
0.1806	45.4545	1500	1.6037	0.7230	0.2878
0.1041	48.4848	1600	1.6310	0.7300	0.2869
0.1451	51.5152	1700	1.6029	0.7275	0.2913
0.132	54.5455	1800	1.7031	0.7444	0.2977
0.1037	57.5758	1900	1.6885	0.7214	0.2822
0.1224	60.6061	2000	1.7125	0.7210	0.2853
0.0921	63.6364	2100	1.7166	0.7333	0.2865
0.1183	66.6667	2200	1.7051	0.7329	0.2800
0.1212	69.6970	2300	1.7752	0.7255	0.2861
0.1153	72.7273	2400	1.7066	0.7279	0.2793
0.0902	75.7576	2500	1.7348	0.7251	0.2838
0.1237	78.7879	2600	1.6664	0.7128	0.2751
0.1001	81.8182	2700	1.7235	0.7103	0.2845
0.0831	84.8485	2800	1.7273	0.7046	0.2847
0.0729	87.8788	2900	1.7377	0.7029	0.2819
0.0755	90.9091	3000	1.7035	0.7107	0.2826
0.0998	93.9394	3100	1.7066	0.6963	0.2793
0.0586	96.9697	3200	1.7023	0.6914	0.2767
0.0669	100.0	3300	1.6979	0.6968	0.2762

Framework versions

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