xlsr-arabic

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: 0.8019
• Wer: 0.5205
• Cer: 0.2091

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: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
9.6085	0.1127	100	8.2716	1.0	1.0
3.7055	0.2255	200	3.5635	1.0	1.0
3.6707	0.3382	300	3.4485	1.0	1.0
3.4696	0.4510	400	3.3925	1.0	0.9896
3.5699	0.5637	500	3.3520	0.9989	0.9810
2.5766	0.6764	600	2.5549	0.9997	0.8222
1.5035	0.7892	700	1.3485	0.9135	0.4360
1.508	0.9019	800	1.0530	0.8492	0.3821
1.094	1.0147	900	0.8783	0.7682	0.3325
0.9471	1.1274	1000	0.8124	0.7197	0.3019
0.8182	1.2401	1100	0.8092	0.7173	0.2788
0.8558	1.3529	1200	0.7785	0.7034	0.2655
0.7199	1.4656	1300	0.6794	0.6689	0.2554
0.7401	1.5784	1400	0.6849	0.6593	0.2509
0.8892	1.6911	1500	0.6862	0.6533	0.2456
0.6824	1.8038	1600	0.6654	0.6426	0.2455
0.5671	1.9166	1700	0.6851	0.6582	0.2378
0.5298	2.0293	1800	0.7284	0.6530	0.2476
0.5056	2.1421	1900	0.6453	0.6348	0.2311
0.4889	2.2548	2000	0.6641	0.6365	0.2342
0.6665	2.3675	2100	0.6607	0.6240	0.2305
0.4022	2.4803	2200	0.6029	0.6054	0.2255
0.5083	2.5930	2300	0.5710	0.5894	0.2159
0.5413	2.7057	2400	0.5762	0.5981	0.2219
0.5765	2.8185	2500	0.5684	0.5965	0.2232
0.6379	2.9312	2600	0.5478	0.5692	0.2121
0.388	3.0440	2700	0.5589	0.5971	0.2220
0.5047	3.1567	2800	0.5903	0.5882	0.2155
0.4911	3.2694	2900	0.5813	0.5838	0.2240
0.4059	3.3822	3000	0.5796	0.5884	0.2208
0.4182	3.4949	3100	0.6368	0.5939	0.2243
0.425	3.6077	3200	0.5325	0.5437	0.2093
0.5876	3.7204	3300	0.5463	0.5629	0.2091
0.3795	3.8331	3400	0.5265	0.5554	0.2090
0.5567	3.9459	3500	0.5372	0.5577	0.2108
0.4698	4.0586	3600	0.5723	0.5900	0.2197
0.3856	4.1714	3700	0.5992	0.5753	0.2168
0.427	4.2841	3800	0.5735	0.5790	0.2156
0.3449	4.3968	3900	0.5642	0.5750	0.2113
0.4049	4.5096	4000	0.5972	0.5825	0.2203
0.4687	4.6223	4100	0.5649	0.5612	0.2111
0.4301	4.7351	4200	0.5515	0.5622	0.2105
0.4429	4.8478	4300	0.5622	0.5709	0.2125
0.4234	4.9605	4400	0.5684	0.5496	0.2098
0.3361	5.0733	4500	0.6108	0.5541	0.2137
0.3547	5.1860	4600	0.5869	0.5508	0.2091
0.2801	5.2988	4700	0.6526	0.5586	0.2163
0.3237	5.4115	4800	0.6481	0.5576	0.2169
0.3366	5.5242	4900	0.5603	0.5345	0.2076
0.2724	5.6370	5000	0.6141	0.5491	0.2151
0.2845	5.7497	5100	0.7205	0.5605	0.2266
0.293	5.8625	5200	0.6246	0.5502	0.2142
0.2904	5.9752	5300	0.5936	0.5386	0.2097
0.3082	6.0879	5400	0.6173	0.5134	0.2032
0.35	6.2007	5500	0.6430	0.5158	0.2051
0.2101	6.3134	5600	0.5861	0.5110	0.1998
0.2822	6.4262	5700	0.6322	0.5269	0.2092
0.263	6.5389	5800	0.7677	0.5477	0.2231
0.2329	6.6516	5900	0.6837	0.5336	0.2129
0.2626	6.7644	6000	0.6350	0.5208	0.2075
0.2467	6.8771	6100	0.6082	0.5274	0.2060
0.3242	6.9899	6200	0.6619	0.5347	0.2098
0.3301	7.1026	6300	0.6798	0.5255	0.2107
0.3085	7.2153	6400	0.6934	0.5202	0.2076
0.3395	7.3281	6500	0.6981	0.5329	0.2125
0.2766	7.4408	6600	0.6886	0.5256	0.2091
0.2479	7.5536	6700	0.7543	0.5414	0.2148
0.18	7.6663	6800	0.7538	0.5198	0.2127
0.3539	7.7790	6900	0.6877	0.5290	0.2136
0.2759	7.8918	7000	0.6516	0.5110	0.2053
0.1152	8.0045	7100	0.7376	0.5293	0.2143
0.1814	8.1172	7200	0.7046	0.5156	0.2068
0.1829	8.2300	7300	0.7658	0.5190	0.2108
0.1165	8.3427	7400	0.8318	0.5210	0.2139
0.1255	8.4555	7500	0.7769	0.5188	0.2085
0.1013	8.5682	7600	0.7409	0.5153	0.2044
0.1273	8.6809	7700	0.7661	0.5181	0.2076
0.1178	8.7937	7800	0.8007	0.5218	0.2113
0.1028	8.9064	7900	0.7513	0.5137	0.2075
0.2003	9.0192	8000	0.7449	0.5133	0.2077
0.1495	9.1319	8100	0.8426	0.5140	0.2105
0.1283	9.2446	8200	0.7653	0.5112	0.2066
0.0585	9.3574	8300	0.7894	0.5176	0.2092
0.1543	9.4701	8400	0.7675	0.5147	0.2064
0.144	9.5829	8500	0.7927	0.5187	0.2096
0.1185	9.6956	8600	0.8045	0.5201	0.2101
0.1707	9.8083	8700	0.7941	0.5193	0.2089
0.0927	9.9211	8800	0.8019	0.5205	0.2091

Framework versions

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