wav2vec2-xls-r-300m-tr

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_11_0 - TR dataset. It achieves the following results on the evaluation set:

• Loss: 0.3179
• Wer: 0.2863
• Cer: 0.0681

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: 64
• 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: 500
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.71	400	1.7290	0.9804	0.4797
4.5435	1.42	800	0.4810	0.5774	0.1450
0.523	2.12	1200	0.3859	0.4812	0.1156
0.3449	2.83	1600	0.3492	0.4498	0.1095
0.2814	3.54	2000	0.3660	0.4466	0.1099
0.2814	4.25	2400	0.3766	0.4235	0.1043
0.2463	4.96	2800	0.3416	0.4119	0.1010
0.2296	5.66	3200	0.3322	0.4013	0.0979
0.2143	6.37	3600	0.3370	0.3956	0.0972
0.1955	7.08	4000	0.3401	0.4033	0.0998
0.1955	7.79	4400	0.3375	0.3889	0.0962
0.1845	8.5	4800	0.3455	0.3752	0.0923
0.1752	9.2	5200	0.3336	0.3718	0.0925
0.1705	9.91	5600	0.3145	0.3653	0.0892
0.1585	10.62	6000	0.3410	0.3737	0.0922
0.1585	11.33	6400	0.3296	0.3664	0.0899
0.1474	12.04	6800	0.3492	0.3590	0.0899
0.1485	12.74	7200	0.3176	0.3506	0.0867
0.137	13.45	7600	0.3532	0.3600	0.0890
0.1291	14.16	8000	0.3318	0.3571	0.0873
0.1291	14.87	8400	0.3353	0.3548	0.0883
0.1274	15.58	8800	0.3235	0.3396	0.0823
0.1198	16.28	9200	0.3259	0.3389	0.0832
0.1164	16.99	9600	0.3263	0.3411	0.0844
0.1119	17.7	10000	0.3254	0.3377	0.0824
0.1119	18.41	10400	0.3243	0.3331	0.0812
0.1054	19.12	10800	0.3223	0.3239	0.0790
0.1017	19.82	11200	0.3054	0.3190	0.0774
0.0964	20.53	11600	0.3278	0.3237	0.0785
0.0903	21.24	12000	0.3167	0.3177	0.0774
0.0903	21.95	12400	0.3331	0.3124	0.0766
0.0886	22.65	12800	0.3099	0.3089	0.0745
0.0836	23.36	13200	0.3171	0.3048	0.0731
0.0796	24.07	13600	0.3158	0.3041	0.0733
0.0739	24.78	14000	0.3203	0.3003	0.0721
0.0739	25.49	14400	0.3138	0.2974	0.0713
0.0742	26.19	14800	0.3197	0.2959	0.0711
0.07	26.9	15200	0.3232	0.2952	0.0703
0.0654	27.61	15600	0.3243	0.2939	0.0701
0.0631	28.32	16000	0.3213	0.2876	0.0688
0.0631	29.03	16400	0.3151	0.2880	0.0685
0.0607	29.73	16800	0.3184	0.2867	0.0681

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.8.1.dev0
• Tokenizers 0.13.2