ssc-led-mms-model-mix-adapt-max3-devtrain

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

• Loss: 0.3148
• Cer: 0.0905
• Wer: 0.2510

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
0.3221	0.2683	200	0.3254	0.0958	0.2812
0.3633	0.5366	400	0.3317	0.0969	0.2750
0.3513	0.8048	600	0.3423	0.0981	0.2753
0.3447	1.0724	800	0.3362	0.0941	0.2638
0.3403	1.3407	1000	0.3415	0.0985	0.2782
0.3358	1.6090	1200	0.3332	0.0954	0.2685
0.3146	1.8773	1400	0.3357	0.0977	0.2825
0.2648	2.1449	1600	0.3264	0.0938	0.2657
0.2934	2.4131	1800	0.3279	0.0953	0.2756
0.3001	2.6814	2000	0.3242	0.0948	0.2671
0.2749	2.9497	2200	0.3238	0.0928	0.2613
0.2686	3.2173	2400	0.3229	0.0925	0.2604
0.2461	3.4856	2600	0.3239	0.0934	0.2642
0.2485	3.7539	2800	0.3190	0.0923	0.2587
0.2703	4.0215	3000	0.3172	0.0909	0.2542
0.2166	4.2897	3200	0.3192	0.0899	0.2497
0.2312	4.5580	3400	0.3170	0.0902	0.2504
0.2244	4.8263	3600	0.3148	0.0905	0.2510

Framework versions

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