wav2vec2-large-mms-1b-hindi-colab

This model is a fine-tuned version of facebook/mms-1b-fl102 on the common_voice_6_1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3535
• Wer: 0.3202

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: 0.001
• train_batch_size: 4
• 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: 10
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
16.7585	0.14	10	10.2106	2.0951
6.9602	0.27	20	3.7700	1.0046
2.4653	0.41	30	1.3321	0.6763
1.0919	0.55	40	0.6594	0.4664
0.7645	0.68	50	0.4930	0.3910
0.8434	0.82	60	0.4819	0.3898
0.5118	0.96	70	0.4492	0.3817
0.6097	1.1	80	0.4299	0.4327
0.4698	1.23	90	0.4308	0.3643
0.5402	1.37	100	0.4042	0.4107
0.5622	1.51	110	0.4156	0.3701
0.4084	1.64	120	0.4138	0.3701
0.4888	1.78	130	0.3917	0.3434
0.4253	1.92	140	0.3852	0.3457
0.5004	2.05	150	0.3843	0.3364
0.3791	2.19	160	0.3841	0.3469
0.3302	2.33	170	0.3764	0.3271
0.4047	2.47	180	0.3689	0.3364
0.2951	2.6	190	0.3657	0.3329
0.3545	2.74	200	0.3582	0.3306
0.3736	2.88	210	0.3585	0.3248
0.388	3.01	220	0.3602	0.3237
0.2997	3.15	230	0.3624	0.3167
0.3704	3.29	240	0.3625	0.3190
0.2095	3.42	250	0.3571	0.3248
0.3564	3.56	260	0.3570	0.3202
0.2119	3.7	270	0.3550	0.3225
0.3697	3.84	280	0.3542	0.3190
0.3551	3.97	290	0.3535	0.3202

Framework versions

• Transformers 4.31.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3