ssc-sco-mms-model-mix-adapt-max

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: 0.5079
• Cer: 0.1382
• Wer: 0.3974

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: 8
• eval_batch_size: 8
• 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
1.1023	0.3419	200	0.7274	0.1928	0.5025
0.9531	0.6838	400	0.6376	0.1665	0.4613
0.8493	1.0256	600	0.5990	0.1583	0.4471
0.8849	1.3675	800	0.5687	0.1531	0.4267
0.8303	1.7094	1000	0.5544	0.1481	0.4195
0.8295	2.0513	1200	0.5455	0.1472	0.4177
0.8169	2.3932	1400	0.5352	0.1465	0.4181
0.7759	2.7350	1600	0.5332	0.1453	0.4129
0.838	3.0769	1800	0.5252	0.1420	0.4042
0.7476	3.4188	2000	0.5220	0.1415	0.4049
0.7028	3.7607	2200	0.5161	0.1404	0.4071
0.7355	4.1026	2400	0.5146	0.1408	0.3999
0.7489	4.4444	2600	0.5122	0.1399	0.4004
0.7108	4.7863	2800	0.5079	0.1382	0.3974

Framework versions

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