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

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

• Loss: 0.1796
• Cer: 0.0677
• Wer: 0.3825

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: 1
• eval_batch_size: 6
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 2
• 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
0.2723	0.2789	200	0.1972	0.0708	0.4000
0.2445	0.5579	400	0.1970	0.0754	0.4108
0.2361	0.8368	600	0.1942	0.0744	0.4055
0.2245	1.1158	800	0.1927	0.0709	0.3989
0.2219	1.3947	1000	0.1930	0.0711	0.4044
0.2404	1.6736	1200	0.1862	0.0698	0.3934
0.2779	1.9526	1400	0.1840	0.0696	0.3910
0.2132	2.2315	1600	0.1864	0.0693	0.3927
0.2173	2.5105	1800	0.1811	0.0689	0.3861
0.2019	2.7894	2000	0.1850	0.0695	0.3863
0.2098	3.0683	2200	0.1819	0.0680	0.3851
0.196	3.3473	2400	0.1836	0.0682	0.3892
0.2333	3.6262	2600	0.1811	0.0682	0.3870
0.2336	3.9052	2800	0.1808	0.0679	0.3837
0.2226	4.1841	3000	0.1806	0.0678	0.3831
0.2016	4.4630	3200	0.1793	0.0677	0.3831
0.1872	4.7420	3400	0.1796	0.0677	0.3825

Framework versions

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