ssc-ady-mms-model-mix-adapt-max2

This model is a fine-tuned version of facebook/mms-1b-all on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.5688
• Cer: 0.3115
• Wer: 1.0067

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: 12
• eval_batch_size: 12
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 24
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.6939	0.4132	200	0.5240	0.1948	0.8809
0.5164	0.8264	400	0.4230	0.1689	0.8082
0.4719	1.2397	600	0.3680	0.1512	0.7403
0.4443	1.6529	800	0.3637	0.1481	0.7297
0.4258	2.0661	1000	0.3406	0.1422	0.6982
0.4051	2.4793	1200	0.3356	0.1420	0.7092
0.3987	2.8926	1400	0.3173	0.1385	0.6826
0.3807	3.3058	1600	0.3281	0.1395	0.6910
0.3725	3.7190	1800	0.3255	0.1399	0.6867
0.3896	4.1322	2000	0.3136	0.1370	0.6771
0.3729	4.5455	2200	0.3116	0.1338	0.6642
0.3729	4.9587	2400	0.3186	0.1360	0.6683
0.3557	5.3719	2600	0.3153	0.1379	0.6728
0.3763	5.7851	2800	0.3302	0.1393	0.6819
0.3988	6.1983	3000	0.3593	0.1407	0.7120
0.4476	6.6116	3200	0.3716	0.1467	0.7161
0.457	7.0248	3400	0.3781	0.1428	0.7013
0.4784	7.4380	3600	0.4148	0.1444	0.7197
0.7014	7.8512	3800	0.5790	0.1576	0.7525
1.2074	8.2645	4000	1.2445	0.2668	0.9548
1.679	8.6777	4200	1.5282	0.4731	1.0718
1.7237	9.0909	4400	1.4973	0.4396	1.0677
1.6836	9.5041	4600	1.4758	0.3751	1.0416
1.7748	9.9174	4800	1.5688	0.3115	1.0067

Framework versions

• Transformers 4.52.1
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4