xlsr-am-adap-ru

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: 1.7523
• Wer: 0.7058
• Cer: 0.2819

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
4.7311	3.0303	100	4.5737	1.0	1.0
4.1724	6.0606	200	4.1384	1.0	0.9880
4.0604	9.0909	300	4.0653	1.0	0.9863
1.8268	12.1212	400	2.0440	0.9610	0.5231
0.9795	15.1515	500	1.6829	0.8855	0.4220
0.8934	18.1818	600	1.4791	0.8375	0.3794
0.4444	21.2121	700	1.4552	0.8174	0.3440
0.3345	24.2424	800	1.3845	0.7862	0.3075
0.3854	27.2727	900	1.5174	0.7846	0.3254
0.2771	30.3030	1000	1.4851	0.7743	0.3161
0.2124	33.3333	1100	1.5143	0.7411	0.3079
0.2229	36.3636	1200	1.5672	0.7468	0.3098
0.2108	39.3939	1300	1.6140	0.7595	0.3094
0.2062	42.4242	1400	1.5804	0.7485	0.3111
0.1375	45.4545	1500	1.6814	0.7489	0.3040
0.1544	48.4848	1600	1.6285	0.7464	0.3030
0.1588	51.5152	1700	1.7064	0.7349	0.3010
0.1559	54.5455	1800	1.7128	0.7485	0.3087
0.1186	57.5758	1900	1.7004	0.7349	0.3004
0.1054	60.6061	2000	1.7703	0.7296	0.2981
0.1306	63.6364	2100	1.7435	0.7304	0.2921
0.1053	66.6667	2200	1.7505	0.7300	0.2951
0.1228	69.6970	2300	1.7204	0.7181	0.2865
0.1359	72.7273	2400	1.6765	0.7173	0.2889
0.0904	75.7576	2500	1.6989	0.7173	0.2831
0.1258	78.7879	2600	1.7484	0.7128	0.2844
0.097	81.8182	2700	1.7627	0.7046	0.2835
0.0814	84.8485	2800	1.7628	0.7165	0.2888
0.0834	87.8788	2900	1.7520	0.7136	0.2886
0.0658	90.9091	3000	1.7474	0.7091	0.2901
0.0977	93.9394	3100	1.7767	0.7066	0.2812
0.0644	96.9697	3200	1.7563	0.7050	0.2835
0.0572	100.0	3300	1.7523	0.7058	0.2819

Framework versions

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