ssc-lth-mms-model-mix-adapt-max2

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

• Loss: 0.8766
• Cer: 0.1753
• Wer: 0.4441

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: 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
0.5128	0.2797	200	0.9431	0.1827	0.4718
0.4979	0.5594	400	0.8099	0.1812	0.4844
0.5554	0.8392	600	0.8049	0.2087	0.5610
0.4722	1.1189	800	0.8562	0.1775	0.4526
0.5403	1.3986	1000	0.8627	0.1826	0.4838
0.495	1.6783	1200	0.8026	0.1833	0.4760
0.483	1.9580	1400	0.8246	0.1772	0.4462
0.4352	2.2378	1600	0.8419	0.1840	0.4835
0.4607	2.5175	1800	0.8474	0.1866	0.4852
0.4196	2.7972	2000	0.8585	0.1802	0.4598
0.3951	3.0769	2200	0.8591	0.1767	0.4574
0.3843	3.3566	2400	0.9117	0.1741	0.4386
0.3912	3.6364	2600	0.8538	0.1806	0.4622
0.4295	3.9161	2800	0.8342	0.1818	0.4665
0.3643	4.1958	3000	0.8846	0.1752	0.4421
0.3686	4.4755	3200	0.8553	0.1791	0.4555
0.3295	4.7552	3400	0.8766	0.1753	0.4441

Framework versions

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