ssc-ush-mms-model-mix-adapt-max-longcv2

This model is a fine-tuned version of facebook/mms-1b-all on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5479
• Cer: 0.1241
• Wer: 0.4626

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.7025	3.125	200	0.5108	0.1573	0.5156
0.4765	6.25	400	0.4266	0.1213	0.4428
0.3969	9.375	600	0.4061	0.1217	0.4158
0.3477	12.5	800	0.4071	0.1241	0.4272
0.3402	15.625	1000	0.3838	0.1132	0.4044
0.3649	18.75	1200	0.4015	0.1206	0.4148
0.6528	21.875	1400	0.6619	0.1336	0.4886
0.5894	25.0	1600	0.4972	0.1197	0.4459
0.5353	28.125	1800	0.5479	0.1241	0.4626

Framework versions

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