ssc-bxk-mms-model-mix-adapt-max3

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.0548
• Cer: 0.1552
• Wer: 0.5916

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: 0.0005
• train_batch_size: 1
• eval_batch_size: 6
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 2
• 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: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.4086	0.2436	200	1.0313	0.1627	0.6095
0.4348	0.4872	400	1.0494	0.1637	0.6129
0.4157	0.7308	600	1.0627	0.1587	0.6298
0.3929	0.9744	800	1.0561	0.1614	0.6097
0.3723	1.2180	1000	1.0516	0.1591	0.6065
0.3728	1.4616	1200	1.0601	0.1623	0.6077
0.3726	1.7052	1400	1.0449	0.1600	0.6052
0.3901	1.9488	1600	1.0422	0.1558	0.5950
0.347	2.1924	1800	1.0450	0.1578	0.5975
0.3592	2.4361	2000	1.0555	0.1577	0.5962
0.3393	2.6797	2200	1.0600	0.1563	0.5975
0.3605	2.9233	2400	1.0558	0.1571	0.5884
0.3416	3.1669	2600	1.0562	0.1563	0.5911
0.352	3.4105	2800	1.0536	0.1572	0.5925
0.3176	3.6541	3000	1.0585	0.1540	0.5881
0.3382	3.8977	3200	1.0541	0.1558	0.5905
0.3137	4.1413	3400	1.0553	0.1559	0.5916
0.3005	4.3849	3600	1.0631	0.1556	0.5913
0.3012	4.6285	3800	1.0547	0.1556	0.5888
0.3264	4.8721	4000	1.0548	0.1552	0.5916

Framework versions

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