ssc-top-mms-model-mix-adapt-max-lowlr

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

• Loss: 0.7049
• Cer: 0.1452
• Wer: 0.5905

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: 8
• eval_batch_size: 6
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.6201	0.3487	200	0.7838	0.1619	0.6789
1.2566	0.6975	400	0.7509	0.1552	0.6508
1.1404	1.0453	600	0.7600	0.1584	0.6621
1.1198	1.3941	800	0.7692	0.1588	0.6555
1.114	1.7428	1000	0.7009	0.1490	0.6146
1.157	2.0907	1200	0.7033	0.1468	0.6002
1.0393	2.4394	1400	0.7133	0.1468	0.5936
1.0582	2.7881	1600	0.7062	0.1450	0.5921
1.0315	3.1360	1800	0.7120	0.1481	0.6084
1.04	3.4847	2000	0.7125	0.1454	0.5897
1.0979	3.8335	2200	0.7051	0.1448	0.5917
0.9606	4.1813	2400	0.7085	0.1451	0.5827
0.9512	4.5301	2600	0.7061	0.1459	0.5893
1.07	4.8788	2800	0.7049	0.1452	0.5905

Framework versions

• Transformers 4.57.2
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.22.0