ssc-ukv-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: 0.5083
• Cer: 0.1348
• Wer: 0.4147

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
0.6983	0.2658	200	0.5624	0.1404	0.4446
0.6518	0.5316	400	0.5353	0.1406	0.4348
0.5823	0.7973	600	0.5431	0.1381	0.4301
0.573	1.0625	800	0.5462	0.1382	0.4358
0.6338	1.3282	1000	0.5299	0.1370	0.4257
0.5693	1.5940	1200	0.5296	0.1369	0.4292
0.5789	1.8598	1400	0.5216	0.1366	0.4249
0.5366	2.1249	1600	0.5204	0.1388	0.4340
0.518	2.3907	1800	0.5237	0.1360	0.4219
0.6573	2.6565	2000	0.5170	0.1346	0.4158
0.557	2.9223	2200	0.5187	0.1356	0.4189
0.5135	3.1874	2400	0.5164	0.1352	0.4182
0.4979	3.4532	2600	0.5103	0.1349	0.4168
0.5908	3.7189	2800	0.5137	0.1350	0.4175
0.553	3.9847	3000	0.5103	0.1351	0.4172
0.5559	4.2498	3200	0.5115	0.1352	0.4184
0.5706	4.5156	3400	0.5094	0.1347	0.4169
0.5647	4.7814	3600	0.5083	0.1348	0.4147

Framework versions

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