xls-r-300m-model-A

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

• Loss: 0.3617
• Wer: 0.3417
• Cer: 0.3359

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
26.6792	1.0	47	10.9107	1.0	1.0
7.8048	2.0	94	3.6963	1.0	1.0
3.4503	3.0	141	3.1623	1.0	1.0
3.1703	4.0	188	3.1486	1.0	1.0
3.1754	5.0	235	3.1398	1.0	1.0
3.19	6.0	282	3.1649	1.0	1.0
3.0893	7.0	329	3.1219	1.0	1.0
2.9698	8.0	376	2.8057	1.0	1.0
2.8622	9.0	423	2.7599	1.0	1.0
2.7293	10.0	470	2.5735	1.0	0.8694
2.6369	11.0	517	2.4812	1.0	0.8518
2.4521	12.0	564	2.3214	1.0	0.7911
2.3405	13.0	611	2.1990	1.0	0.7530
2.1983	14.0	658	2.0527	1.0	0.7339
2.1379	15.0	705	2.0385	1.0	0.7283
2.0842	16.0	752	1.9515	1.0	0.7198
2.0469	17.0	799	1.9015	1.0	0.7205
1.9494	18.0	846	1.9291	1.0	0.7163
2.0306	19.0	893	1.8532	1.0	0.7092
1.8124	20.0	940	1.6457	1.0	0.6733
1.7114	21.0	987	1.4539	0.9561	0.6239
1.5781	22.0	1034	1.5011	0.9592	0.7368
1.486	23.0	1081	1.4219	0.9154	0.5801
1.3247	24.0	1128	1.1263	0.8997	0.5293
1.2121	25.0	1175	1.5609	0.9279	0.6598
1.1214	26.0	1222	0.9734	0.8966	0.4919
0.9949	27.0	1269	1.0702	0.8966	0.5159
0.905	28.0	1316	0.7401	0.8809	0.4263
0.8187	29.0	1363	0.6565	0.8056	0.3924
0.6954	30.0	1410	0.5022	0.7398	0.3578
0.6616	31.0	1457	0.4858	0.7085	0.3634
0.5332	32.0	1504	0.4242	0.6207	0.2844
0.5323	33.0	1551	0.3704	0.5705	0.2583
0.4133	34.0	1598	0.3560	0.5831	0.2682
0.4121	35.0	1645	0.3449	0.5517	0.2696
0.4243	36.0	1692	0.4157	0.5517	0.2802
0.3316	37.0	1739	0.4054	0.5611	0.3507
0.2774	38.0	1786	0.4340	0.5517	0.3606
0.3415	39.0	1833	0.3619	0.5298	0.3409
0.2722	40.0	1880	0.3098	0.4232	0.3705
0.2532	41.0	1927	0.2978	0.3950	0.2569
0.173	42.0	1974	0.2661	0.3605	0.2167
0.2258	43.0	2021	0.2203	0.3480	0.2329
0.1722	44.0	2068	0.3302	0.3605	0.2646
0.1648	45.0	2115	0.2736	0.4107	0.3507
0.2142	46.0	2162	0.4359	0.3762	0.3366
0.1417	47.0	2209	0.3170	0.3417	0.3331
0.1138	48.0	2256	0.2937	0.3448	0.2548
0.1506	49.0	2303	0.3907	0.3636	0.3387
0.0958	50.0	2350	0.3740	0.3448	0.3020
0.1082	51.0	2397	0.2869	0.3135	0.2809
0.1229	52.0	2444	0.2933	0.3542	0.3267
0.1173	53.0	2491	0.3282	0.3480	0.3070
0.1222	54.0	2538	0.3940	0.3542	0.3458
0.0832	55.0	2585	0.3013	0.3354	0.3126
0.0599	56.0	2632	0.3137	0.3386	0.3282
0.0864	57.0	2679	0.3238	0.3386	0.3225
0.0632	58.0	2726	0.3388	0.3417	0.3338
0.0741	59.0	2773	0.3574	0.3417	0.3345
0.0587	60.0	2820	0.3617	0.3417	0.3359

Framework versions

• Transformers 4.28.0
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.13.3