asr-africa/bambara-5-hours-bambara-asr-hf-plus-lm

bambara-5-hours-bambara-asr-hf

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: 1.8042
• Wer: 0.5257
• Cer: 0.2399

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: 16
• 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: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
5.6981	1.13	200	2.9919	1.0	1.0
2.9496	2.27	400	2.9333	1.0	1.0
2.4081	3.4	600	1.4375	0.8869	0.4174
1.5268	4.53	800	1.1942	0.7734	0.3602
1.3291	5.67	1000	1.1261	0.6712	0.2994
1.2301	6.8	1200	1.0859	0.6470	0.2892
1.1379	7.93	1400	1.0552	0.6314	0.2813
1.0664	9.07	1600	1.0684	0.6079	0.2731
0.9866	10.2	1800	1.0584	0.6110	0.2756
0.9625	11.33	2000	1.0776	0.5757	0.2590
0.8828	12.46	2200	1.1372	0.5865	0.2596
0.8451	13.6	2400	1.0821	0.5645	0.2574
0.8016	14.73	2600	1.1293	0.5754	0.2608
0.7615	15.86	2800	1.1312	0.5586	0.2519
0.715	17.0	3000	1.1657	0.5635	0.2558
0.6792	18.13	3200	1.2197	0.5521	0.2517
0.6498	19.26	3400	1.1157	0.5606	0.2533
0.6221	20.4	3600	1.2632	0.5501	0.2507
0.5912	21.53	3800	1.1686	0.5520	0.2505
0.5614	22.66	4000	1.3080	0.5547	0.2542
0.5485	23.8	4200	1.2349	0.5601	0.2583
0.5235	24.93	4400	1.2541	0.5458	0.2489
0.5084	26.06	4600	1.2414	0.5500	0.2514
0.4741	27.2	4800	1.5119	0.5444	0.2470
0.4699	28.33	5000	1.2778	0.5525	0.2524
0.4511	29.46	5200	1.5488	0.5502	0.2494
0.4181	30.59	5400	1.3489	0.5522	0.2536
0.4149	31.73	5600	1.5685	0.5460	0.2454
0.3998	32.86	5800	1.4369	0.5434	0.2494
0.401	33.99	6000	1.5961	0.5376	0.2433
0.3744	35.13	6200	1.5695	0.5361	0.2452
0.36	36.26	6400	1.5968	0.5400	0.2445
0.3435	37.39	6600	1.6238	0.5334	0.2424
0.3337	38.53	6800	1.6619	0.5340	0.2440
0.3232	39.66	7000	1.6444	0.5345	0.2446
0.3224	40.79	7200	1.7175	0.5417	0.2451
0.303	41.93	7400	1.6966	0.5273	0.2417
0.3028	43.06	7600	1.7403	0.5325	0.2431
0.2899	44.19	7800	1.7688	0.5255	0.2398
0.2861	45.33	8000	1.7705	0.5269	0.2405
0.2744	46.46	8200	1.7792	0.5244	0.2396
0.2731	47.59	8400	1.7876	0.5286	0.2412
0.27	48.73	8600	1.8060	0.5264	0.2403
0.2643	49.86	8800	1.8042	0.5257	0.2399

Framework versions

• Transformers 4.38.1
• Pytorch 2.1.0+cu118
• Datasets 2.17.0
• Tokenizers 0.15.2