wav2vec2-large-mms-1b-arabic-colab

This model is a fine-tuned version of facebook/mms-1b-all on the fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.2922
• Wer: 0.2591

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: 1e-05
• train_batch_size: 4
• 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: 100
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
18.9344	0.19	100	17.8048	1.0
15.6959	0.38	200	14.1448	1.0
11.9387	0.57	300	9.8417	1.0
7.554	0.76	400	5.3727	1.0
4.3953	0.95	500	3.5681	1.0
3.3533	1.14	600	3.1439	1.0
2.9309	1.33	700	2.5171	0.9987
2.1985	1.52	800	1.7128	0.8522
1.5126	1.71	900	1.1276	0.5744
1.0376	1.9	1000	0.7830	0.4400
0.7702	2.09	1100	0.5959	0.3765
0.6274	2.28	1200	0.4986	0.3363
0.5423	2.47	1300	0.4473	0.3197
0.494	2.66	1400	0.4153	0.3046
0.4372	2.85	1500	0.3940	0.2946
0.4667	3.04	1600	0.3791	0.2887
0.4228	3.23	1700	0.3670	0.2823
0.4177	3.42	1800	0.3571	0.2803
0.3824	3.61	1900	0.3494	0.2789
0.4002	3.8	2000	0.3435	0.2782
0.4112	3.99	2100	0.3385	0.2776
0.3788	4.18	2200	0.3342	0.2768
0.4079	4.37	2300	0.3305	0.2752
0.3939	4.56	2400	0.3271	0.2733
0.3601	4.75	2500	0.3250	0.2724
0.3443	4.94	2600	0.3223	0.2727
0.3723	5.13	2700	0.3200	0.2724
0.3669	5.32	2800	0.3182	0.2704
0.3117	5.51	2900	0.3167	0.2693
0.3658	5.7	3000	0.3150	0.2694
0.3731	5.89	3100	0.3132	0.2683
0.3542	6.08	3200	0.3122	0.2684
0.3667	6.27	3300	0.3108	0.2681
0.3115	6.46	3400	0.3099	0.2671
0.3466	6.65	3500	0.3092	0.2663
0.3497	6.84	3600	0.3082	0.2656
0.3276	7.03	3700	0.3076	0.2667
0.3316	7.22	3800	0.3070	0.2651
0.3324	7.41	3900	0.3060	0.2656
0.323	7.6	4000	0.3054	0.2661
0.3411	7.79	4100	0.3045	0.2641
0.3583	7.98	4200	0.3037	0.2649
0.3299	8.17	4300	0.3035	0.2649
0.2899	8.37	4400	0.3030	0.2643
0.3432	8.56	4500	0.3025	0.2651
0.3275	8.75	4600	0.3018	0.2631
0.3652	8.94	4700	0.3011	0.2637
0.3373	9.13	4800	0.3009	0.2626
0.3097	9.32	4900	0.3005	0.2627
0.3163	9.51	5000	0.2997	0.2623
0.3443	9.7	5100	0.2995	0.2623
0.346	9.89	5200	0.2989	0.2626
0.302	10.08	5300	0.2988	0.2624
0.3252	10.27	5400	0.2983	0.2623
0.3316	10.46	5500	0.2980	0.2632
0.3424	10.65	5600	0.2975	0.2629
0.3205	10.84	5700	0.2977	0.2622
0.3164	11.03	5800	0.2973	0.2618
0.3348	11.22	5900	0.2968	0.2619
0.3236	11.41	6000	0.2967	0.2612
0.3073	11.6	6100	0.2962	0.2627
0.3129	11.79	6200	0.2964	0.2623
0.3319	11.98	6300	0.2961	0.2621
0.2974	12.17	6400	0.2960	0.2613
0.3557	12.36	6500	0.2955	0.2612
0.3068	12.55	6600	0.2957	0.2619
0.3292	12.74	6700	0.2954	0.2619
0.3278	12.93	6800	0.2952	0.2612
0.314	13.12	6900	0.2948	0.2614
0.3182	13.31	7000	0.2949	0.2618
0.3322	13.5	7100	0.2948	0.2612
0.3089	13.69	7200	0.2944	0.2616
0.3176	13.88	7300	0.2943	0.2613
0.3025	14.07	7400	0.2942	0.2612
0.3277	14.26	7500	0.2941	0.2613
0.3241	14.45	7600	0.2940	0.2617
0.3084	14.64	7700	0.2938	0.2614
0.324	14.83	7800	0.2935	0.2612
0.3229	15.02	7900	0.2934	0.2609
0.3224	15.21	8000	0.2933	0.2602
0.2859	15.4	8100	0.2932	0.2604
0.3173	15.59	8200	0.2931	0.2598
0.3399	15.78	8300	0.2931	0.2602
0.3176	15.97	8400	0.2930	0.2598
0.2993	16.16	8500	0.2930	0.2602
0.3289	16.35	8600	0.2930	0.2598
0.3149	16.54	8700	0.2928	0.2601
0.3172	16.73	8800	0.2927	0.2599
0.3204	16.92	8900	0.2926	0.2597
0.3117	17.11	9000	0.2926	0.2604
0.3051	17.3	9100	0.2927	0.2608
0.3296	17.49	9200	0.2927	0.2604
0.309	17.68	9300	0.2926	0.2602
0.3138	17.87	9400	0.2925	0.2593
0.2802	18.06	9500	0.2925	0.2594
0.308	18.25	9600	0.2925	0.2593
0.3076	18.44	9700	0.2925	0.2591
0.312	18.63	9800	0.2923	0.2592
0.31	18.82	9900	0.2923	0.2593
0.3317	19.01	10000	0.2923	0.2592
0.3357	19.2	10100	0.2923	0.2593
0.302	19.39	10200	0.2922	0.2596
0.294	19.58	10300	0.2923	0.2592
0.3158	19.77	10400	0.2923	0.2593
0.3025	19.96	10500	0.2922	0.2591

Framework versions

• Transformers 4.35.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 2.14.6
• Tokenizers 0.14.1