conformer_wolof-lam-asr

This model is a fine-tuned version of facebook/wav2vec2-conformer-rope-large-960h-ft on the ABDOUAZIIZ/WOLOF_LAM_ASR - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.3302
• Wer: 0.3318

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: 2e-05
• 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_steps: 500
• num_epochs: 32.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.1269	0.5388	1000	2.9698	1.0
0.9518	1.0776	2000	0.6929	0.6198
0.6386	1.6164	3000	0.5234	0.5734
0.5484	2.1552	4000	0.4683	0.4873
0.5115	2.6940	5000	0.4364	0.4393
0.4683	3.2328	6000	0.4189	0.4393
0.4466	3.7716	7000	0.4077	0.3951
0.413	4.3103	8000	0.3967	0.3843
0.4078	4.8491	9000	0.3822	0.3865
0.3772	5.3879	10000	0.3830	0.3585
0.3778	5.9267	11000	0.3640	0.3635
0.3545	6.4655	12000	0.3772	0.3680
0.3433	7.0043	13000	0.3602	0.3594
0.3313	7.5431	14000	0.3518	0.3396
0.3204	8.0819	15000	0.3515	0.3279
0.2983	8.6207	16000	0.3489	0.3243
0.2884	9.1595	17000	0.3394	0.3397
0.2804	9.6983	18000	0.3433	0.3380
0.2623	10.2371	19000	0.3454	0.3340
0.2599	10.7759	20000	0.3378	0.3345
0.2533	11.3147	21000	0.3371	0.3271
0.2489	11.8534	22000	0.3357	0.3260
0.2277	12.3922	23000	0.3302	0.3318
0.2345	12.9310	24000	0.3387	0.3355
0.2146	13.4698	25000	0.3372	0.3220
0.2124	14.0086	26000	0.3335	0.3273

Framework versions

• Transformers 4.41.2
• Pytorch 1.11.0+cu113
• Datasets 2.19.2
• Tokenizers 0.19.1