xlsr-syntesized-turkish-8-hour-llr-2

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: 0.3058
Wer: 0.2940

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: 0.0025
train_batch_size: 2
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 200
num_epochs: 20
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.8595	0.26	100	3.0079	1.0
2.6103	0.52	200	1.6661	1.0
1.0733	0.78	300	0.8518	0.9716
0.7436	1.04	400	0.4824	0.6979
0.5821	1.3	500	0.4184	0.6479
0.5916	1.56	600	0.4041	0.6042
0.5099	1.82	700	0.3637	0.6122
0.4787	2.08	800	0.3393	0.5328
0.4304	2.34	900	0.3145	0.4870
0.4366	2.6	1000	0.3054	0.4718
0.4144	2.86	1100	0.3118	0.4915
0.4303	3.12	1200	0.3303	0.4612
0.3768	3.39	1300	0.3169	0.4844
0.39	3.65	1400	0.2931	0.4351
0.3717	3.91	1500	0.2824	0.4303
0.3396	4.17	1600	0.2606	0.3929
0.341	4.43	1700	0.2845	0.4712
0.3427	4.69	1800	0.2820	0.4452
0.3285	4.95	1900	0.2577	0.3957
0.3011	5.21	2000	0.2486	0.3717
0.2923	5.47	2100	0.2482	0.3990
0.298	5.73	2200	0.2333	0.3858
0.2939	5.99	2300	0.2435	0.3900
0.285	6.25	2400	0.2884	0.4040
0.2763	6.51	2500	0.2518	0.3880
0.2764	6.77	2600	0.2429	0.3832
0.2711	7.03	2700	0.2408	0.3729
0.2457	7.29	2800	0.2270	0.3799
0.2428	7.55	2900	0.2354	0.3761
0.2457	7.81	3000	0.2309	0.3640
0.2413	8.07	3100	0.2408	0.3426
0.224	8.33	3200	0.2233	0.3452
0.2295	8.59	3300	0.2324	0.3320
0.2316	8.85	3400	0.2329	0.3434
0.2156	9.11	3500	0.2506	0.3493
0.2057	9.38	3600	0.2416	0.3475
0.2091	9.64	3700	0.2421	0.3413
0.2105	9.9	3800	0.2557	0.3514
0.2086	10.16	3900	0.2281	0.3365
0.1848	10.42	4000	0.2267	0.3214
0.203	10.68	4100	0.2292	0.3598
0.1863	10.94	4200	0.2242	0.3527
0.1732	11.2	4300	0.2426	0.3359
0.1774	11.46	4400	0.2313	0.3438
0.1699	11.72	4500	0.2326	0.3343
0.171	11.98	4600	0.2360	0.3124
0.1691	12.24	4700	0.2566	0.3322
0.1632	12.5	4800	0.2556	0.2985
0.1637	12.76	4900	0.2536	0.3161
0.1619	13.02	5000	0.2418	0.3493
0.1476	13.28	5100	0.2419	0.3159
0.1542	13.54	5200	0.2742	0.3214
0.1481	13.8	5300	0.2454	0.2971
0.1446	14.06	5400	0.2520	0.2910
0.137	14.32	5500	0.2470	0.2980
0.1365	14.58	5600	0.2563	0.3008
0.1369	14.84	5700	0.2620	0.3033
0.1285	15.1	5800	0.2735	0.2861
0.1266	15.36	5900	0.2816	0.2948
0.1285	15.62	6000	0.2842	0.2890
0.1281	15.89	6100	0.2715	0.2889
0.1264	16.15	6200	0.2779	0.2954
0.1178	16.41	6300	0.2944	0.3113
0.1252	16.67	6400	0.2837	0.2901
0.119	16.93	6500	0.2861	0.3111
0.1203	17.19	6600	0.2785	0.3054
0.1108	17.45	6700	0.2823	0.3101
0.11	17.71	6800	0.2871	0.2926
0.11	17.97	6900	0.2803	0.2995
0.1065	18.23	7000	0.2920	0.2947
0.1047	18.49	7100	0.3047	0.2972
0.1123	18.75	7200	0.2945	0.2978
0.1159	19.01	7300	0.2999	0.2976
0.1061	19.27	7400	0.2954	0.2945
0.1011	19.53	7500	0.3024	0.2969
0.1007	19.79	7600	0.3058	0.2940

Framework versions

Transformers 4.26.0
Pytorch 2.1.0+cu118
Datasets 2.9.0
Tokenizers 0.13.3

alikanakar/xlsr-syntesized-turkish-8-hour-llr-2

xlsr-syntesized-turkish-8-hour-llr-2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

为此模型做种