wav2vec-turkish

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_16_1 dataset. It achieves the following results on the evaluation set:

Loss: 0.3312
Wer: 0.2795

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
5.433	0.29	400	1.1787	0.9025
0.7193	0.58	800	0.6239	0.6505
0.5243	0.88	1200	0.5098	0.5901
0.4514	1.17	1600	0.4618	0.5131
0.419	1.46	2000	0.4341	0.4990
0.3975	1.75	2400	0.4016	0.4809
0.3756	2.05	2800	0.3926	0.4684
0.3421	2.34	3200	0.3841	0.4639
0.3418	2.63	3600	0.3889	0.4551
0.3409	2.92	4000	0.3615	0.4295
0.3039	3.21	4400	0.3939	0.4562
0.2934	3.51	4800	0.3866	0.4531
0.2971	3.8	5200	0.3891	0.4497
0.2953	4.09	5600	0.3694	0.4405
0.2836	4.38	6000	0.3583	0.4252
0.2721	4.67	6400	0.3562	0.4164
0.2685	4.97	6800	0.3574	0.4215
0.251	5.26	7200	0.3660	0.4239
0.2537	5.55	7600	0.3723	0.4308
0.2629	5.84	8000	0.3758	0.4359
0.2469	6.14	8400	0.3799	0.4295
0.2342	6.43	8800	0.3453	0.3947
0.2306	6.72	9200	0.3361	0.3977
0.2284	7.01	9600	0.3592	0.3970
0.213	7.3	10000	0.3451	0.3904
0.2188	7.6	10400	0.3426	0.3828
0.2239	7.89	10800	0.3392	0.3878
0.205	8.18	11200	0.3729	0.4021
0.2049	8.47	11600	0.3511	0.3981
0.2082	8.77	12000	0.3719	0.4143
0.2047	9.06	12400	0.3569	0.3984
0.1895	9.35	12800	0.3416	0.3798
0.1935	9.64	13200	0.3378	0.3793
0.1963	9.93	13600	0.3301	0.3883
0.1889	10.23	14000	0.3577	0.3881
0.182	10.52	14400	0.3281	0.3776
0.1794	10.81	14800	0.3368	0.3780
0.1736	11.1	15200	0.3275	0.3664
0.1659	11.4	15600	0.3504	0.3753
0.1651	11.69	16000	0.3343	0.3733
0.1735	11.98	16400	0.3510	0.3750
0.1569	12.27	16800	0.3243	0.3558
0.1535	12.56	17200	0.3239	0.3603
0.1588	12.86	17600	0.3372	0.3655
0.1524	13.15	18000	0.3453	0.3709
0.1453	13.44	18400	0.3301	0.3590
0.1483	13.73	18800	0.3443	0.3597
0.1432	14.02	19200	0.3401	0.3584
0.1374	14.32	19600	0.3357	0.3618
0.1399	14.61	20000	0.3386	0.3621
0.142	14.9	20400	0.3136	0.3547
0.1307	15.19	20800	0.3328	0.3501
0.1299	15.49	21200	0.3346	0.3458
0.1301	15.78	21600	0.3188	0.3473
0.1285	16.07	22000	0.3323	0.3522
0.1197	16.36	22400	0.3333	0.3392
0.1225	16.65	22800	0.3545	0.3590
0.1263	16.95	23200	0.3360	0.3410
0.1134	17.24	23600	0.3204	0.3332
0.114	17.53	24000	0.3264	0.3349
0.1165	17.82	24400	0.3160	0.3323
0.1134	18.12	24800	0.3479	0.3377
0.1066	18.41	25200	0.3306	0.3378
0.1027	18.7	25600	0.3286	0.3286
0.1083	18.99	26000	0.3285	0.3227
0.0937	19.28	26400	0.3240	0.3259
0.1007	19.58	26800	0.3286	0.3283
0.0996	19.87	27200	0.3278	0.3277
0.0972	20.16	27600	0.3171	0.3212
0.0927	20.45	28000	0.3426	0.3283
0.0932	20.75	28400	0.3418	0.3215
0.0932	21.04	28800	0.3246	0.3192
0.086	21.33	29200	0.3385	0.3201
0.0868	21.62	29600	0.3441	0.3164
0.0875	21.91	30000	0.3246	0.3161
0.0815	22.21	30400	0.3303	0.3105
0.0832	22.5	30800	0.3288	0.3062
0.0781	22.79	31200	0.3411	0.3098
0.077	23.08	31600	0.3343	0.3146
0.0755	23.37	32000	0.3211	0.3093
0.0742	23.67	32400	0.3268	0.3044
0.0721	23.96	32800	0.3222	0.3045
0.0699	24.25	33200	0.3266	0.2993
0.0663	24.54	33600	0.3410	0.3008
0.0719	24.84	34000	0.3221	0.3014
0.0682	25.13	34400	0.3290	0.2976
0.0674	25.42	34800	0.3356	0.2967
0.0661	25.71	35200	0.3181	0.2964
0.0681	26.0	35600	0.3318	0.2964
0.0619	26.3	36000	0.3220	0.2945
0.0617	26.59	36400	0.3270	0.2913
0.0592	26.88	36800	0.3391	0.2909
0.0569	27.17	37200	0.3394	0.2900
0.0557	27.47	37600	0.3359	0.2877
0.0555	27.76	38000	0.3306	0.2847
0.055	28.05	38400	0.3344	0.2827
0.0516	28.34	38800	0.3389	0.2845
0.0544	28.63	39200	0.3360	0.2840
0.0542	28.93	39600	0.3366	0.2828
0.0524	29.22	40000	0.3343	0.2819
0.0527	29.51	40400	0.3319	0.2803
0.0503	29.8	40800	0.3312	0.2795

Framework versions

Transformers 4.38.1
Pytorch 2.2.0+cu121
Datasets 2.17.0
Tokenizers 0.15.2

tgrhn/wav2vec-turkish-300m

wav2vec-turkish

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

为此模型做种