badili/metaasr-1-b-mcv-11-v-1

wav2vec2-large-xls-r-1b-swahili-v12

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4658
• Wer: 0.2038

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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.726	0.35	400	0.7214	0.6674
0.5241	0.69	800	0.5641	0.5345
0.4616	1.04	1200	0.5112	0.4755
0.4018	1.39	1600	0.4797	0.4158
0.3916	1.74	2000	0.4483	0.3985
0.3661	2.08	2400	0.4449	0.3931
0.3314	2.43	2800	0.4124	0.3549
0.3287	2.78	3200	0.4008	0.3651
0.317	3.13	3600	0.4460	0.3735
0.3026	3.47	4000	0.4165	0.3753
0.3061	3.82	4400	0.4112	0.3550
0.2808	4.17	4800	0.3951	0.3275
0.2641	4.52	5200	0.3934	0.3340
0.2709	4.86	5600	0.3963	0.3287
0.2586	5.21	6000	0.4114	0.3396
0.2487	5.56	6400	0.3821	0.3214
0.2618	5.91	6800	0.3987	0.3268
0.2297	6.25	7200	0.3810	0.3132
0.2337	6.6	7600	0.3740	0.3131
0.2285	6.95	8000	0.3715	0.3093
0.2173	7.29	8400	0.3878	0.3147
0.2251	7.64	8800	0.3862	0.3134
0.2215	7.99	9200	0.3621	0.2940
0.195	8.34	9600	0.3651	0.3005
0.201	8.68	10000	0.3837	0.3167
0.1964	9.03	10400	0.3719	0.2876
0.1741	9.38	10800	0.3637	0.2840
0.181	9.73	11200	0.3616	0.2914
0.1795	10.07	11600	0.3719	0.2753
0.1602	10.42	12000	0.3618	0.2856
0.1753	10.77	12400	0.3570	0.2788
0.1627	11.12	12800	0.3500	0.2719
0.1566	11.46	13200	0.3553	0.2808
0.1589	11.81	13600	0.3635	0.2699
0.1511	12.16	14000	0.3656	0.2692
0.1451	12.51	14400	0.3759	0.2759
0.1444	12.85	14800	0.3607	0.2677
0.1359	13.2	15200	0.3852	0.2660
0.1313	13.55	15600	0.3587	0.2679
0.1329	13.89	16000	0.3548	0.2584
0.1163	14.24	16400	0.3701	0.2535
0.1175	14.59	16800	0.3693	0.2638
0.1242	14.94	17200	0.3660	0.2565
0.1067	15.28	17600	0.3835	0.2581
0.1077	15.63	18000	0.3799	0.2504
0.1099	15.98	18400	0.3598	0.2478
0.0952	16.33	18800	0.3865	0.2563
0.1007	16.67	19200	0.3630	0.2565
0.0999	17.02	19600	0.3912	0.2505
0.0895	17.37	20000	0.3934	0.2631
0.0974	17.72	20400	0.3718	0.2462
0.0939	18.06	20800	0.4001	0.2587
0.0915	18.41	21200	0.4048	0.2468
0.0865	18.76	21600	0.3860	0.2415
0.0784	19.11	22000	0.4148	0.2454
0.0782	19.45	22400	0.3952	0.2471
0.0775	19.8	22800	0.3943	0.2434
0.0735	20.15	23200	0.4093	0.2405
0.0679	20.5	23600	0.3996	0.2362
0.0677	20.84	24000	0.4133	0.2365
0.0687	21.19	24400	0.4303	0.2330
0.0651	21.54	24800	0.4288	0.2326
0.0647	21.88	25200	0.4134	0.2347
0.0634	22.23	25600	0.4148	0.2312
0.0592	22.58	26000	0.4322	0.2315
0.06	22.93	26400	0.4050	0.2313
0.0561	23.27	26800	0.4260	0.2263
0.0546	23.62	27200	0.4228	0.2238
0.0548	23.97	27600	0.4140	0.2258
0.0505	24.32	28000	0.4304	0.2246
0.0501	24.66	28400	0.4241	0.2233
0.0481	25.01	28800	0.4385	0.2209
0.0469	25.36	29200	0.4451	0.2189
0.0464	25.71	29600	0.4397	0.2217
0.0438	26.05	30000	0.4419	0.2154
0.0432	26.4	30400	0.4366	0.2137
0.0419	26.75	30800	0.4371	0.2137
0.0419	27.1	31200	0.4552	0.2109
0.0392	27.44	31600	0.4496	0.2108
0.0386	27.79	32000	0.4585	0.2096
0.0387	28.14	32400	0.4496	0.2065
0.0367	28.48	32800	0.4646	0.2082
0.0357	28.83	33200	0.4553	0.2067
0.0355	29.18	33600	0.4615	0.2055
0.0345	29.53	34000	0.4670	0.2046
0.0346	29.87	34400	0.4658	0.2038

Framework versions

• Transformers 4.29.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.12.0
• Tokenizers 0.13.3