wav2vec2-large-xls-r-300m-urdu-cv-10

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_10_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5959
• Wer: 0.3946

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: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
20.8724	0.25	32	18.0006	1.0
10.984	0.5	64	6.8001	1.0
5.7792	0.74	96	4.9273	1.0
4.2891	0.99	128	3.8379	1.0
3.4937	1.24	160	3.2877	1.0
3.1605	1.49	192	3.1198	1.0
3.0874	1.74	224	3.0542	1.0
3.0363	1.98	256	3.0063	0.9999
2.9776	2.23	288	2.9677	1.0
2.8168	2.48	320	2.4189	1.0000
2.0575	2.73	352	1.5330	0.8520
1.4248	2.98	384	1.1747	0.7519
1.1354	3.22	416	0.9837	0.7047
1.0049	3.47	448	0.9414	0.6631
0.956	3.72	480	0.8948	0.6606
0.8906	3.97	512	0.8381	0.6291
0.7587	4.22	544	0.7714	0.5898
0.7534	4.47	576	0.8237	0.5908
0.7203	4.71	608	0.7731	0.5758
0.6876	4.96	640	0.7467	0.5390
0.5825	5.21	672	0.6940	0.5401
0.5565	5.46	704	0.6826	0.5248
0.5598	5.71	736	0.6387	0.5204
0.5289	5.95	768	0.6432	0.4956
0.4565	6.2	800	0.6643	0.4876
0.4576	6.45	832	0.6295	0.4758
0.4265	6.7	864	0.6227	0.4673
0.4359	6.95	896	0.6077	0.4598
0.3576	7.19	928	0.5800	0.4477
0.3612	7.44	960	0.5837	0.4500
0.345	7.69	992	0.5892	0.4466
0.3707	7.94	1024	0.6217	0.4380
0.3269	8.19	1056	0.5964	0.4412
0.2974	8.43	1088	0.6116	0.4394
0.2932	8.68	1120	0.5764	0.4235
0.2854	8.93	1152	0.5757	0.4239
0.2651	9.18	1184	0.5798	0.4253
0.2508	9.43	1216	0.5750	0.4316
0.238	9.67	1248	0.6038	0.4232
0.2454	9.92	1280	0.5781	0.4078
0.2196	10.17	1312	0.5931	0.4178
0.2036	10.42	1344	0.6134	0.4116
0.2087	10.67	1376	0.5831	0.4146
0.1908	10.91	1408	0.5987	0.4159
0.1751	11.16	1440	0.5968	0.4065
0.1726	11.41	1472	0.6037	0.4119
0.1728	11.66	1504	0.5961	0.4011
0.1772	11.91	1536	0.5903	0.3972
0.1647	12.16	1568	0.5960	0.4024
0.1506	12.4	1600	0.5986	0.3933
0.1383	12.65	1632	0.5893	0.3938
0.1433	12.9	1664	0.5999	0.3975
0.1356	13.15	1696	0.6035	0.3982
0.1431	13.4	1728	0.5997	0.4042
0.1346	13.64	1760	0.6018	0.4003
0.1363	13.89	1792	0.5891	0.3969
0.1323	14.14	1824	0.5983	0.3925
0.1196	14.39	1856	0.6003	0.3939
0.1266	14.64	1888	0.5997	0.3941
0.1269	14.88	1920	0.5959	0.3946

Framework versions

• Transformers 4.21.1
• Pytorch 1.11.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1