xls-r-300m-hbs-pl-unfrozen-batch16

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: 0.7148
• Wer: 0.4234
• Cer: 0.0987

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
3.4468	3.2258	100	3.3016	1.0	1.0
3.2215	6.4516	200	3.2110	1.0	1.0
0.58	9.6774	300	0.6904	0.6797	0.1704
0.2905	12.9032	400	0.6428	0.6160	0.1557
0.1415	16.1290	500	0.6332	0.5403	0.1318
0.117	19.3548	600	0.6575	0.5307	0.1286
0.0839	22.5806	700	0.6940	0.5164	0.1256
0.0833	25.8065	800	0.6665	0.4906	0.1176
0.0649	29.0323	900	0.6755	0.4775	0.1148
0.0547	32.2581	1000	0.7033	0.4918	0.1173
0.076	35.4839	1100	0.7090	0.4738	0.1144
0.0505	38.7097	1200	0.7064	0.4756	0.1141
0.0262	41.9355	1300	0.6809	0.4667	0.1126
0.0253	45.1613	1400	0.7226	0.4672	0.1125
0.0467	48.3871	1500	0.7337	0.4644	0.1125
0.0438	51.6129	1600	0.7645	0.4667	0.1101
0.0339	54.8387	1700	0.7208	0.4494	0.1078
0.0568	58.0645	1800	0.7380	0.4534	0.1091
0.0231	61.2903	1900	0.7438	0.4557	0.1091
0.0686	64.5161	2000	0.6985	0.4510	0.1068
0.0276	67.7419	2100	0.7458	0.4492	0.1065
0.0408	70.9677	2200	0.7562	0.4508	0.1077
0.0214	74.1935	2300	0.7325	0.4482	0.1070
0.0116	77.4194	2400	0.7260	0.4388	0.1034
0.0215	80.6452	2500	0.7117	0.4344	0.1028
0.0321	83.8710	2600	0.7227	0.4290	0.1007
0.0236	87.0968	2700	0.7164	0.4276	0.1015
0.0245	90.3226	2800	0.7106	0.4297	0.1013
0.023	93.5484	2900	0.7092	0.4227	0.0990
0.0392	96.7742	3000	0.7176	0.4239	0.0983
0.0073	100.0	3100	0.7148	0.4234	0.0987

Framework versions

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