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

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: 1.7505
• Cer: 0.6135
• Wer: 0.9685

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: 8
• eval_batch_size: 6
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
2.0262	3.2810	200	1.4097	0.1814	0.6275
1.5492	6.5620	400	1.3460	0.1539	0.5110
1.3294	9.8430	600	1.2562	0.1477	0.4753
1.0855	13.1157	800	1.2973	0.1402	0.4680
1.219	16.3967	1000	1.2326	0.1508	0.4995
1.5589	19.6777	1200	1.9629	0.7455	0.9811
2.3588	22.9587	1400	2.3892	0.3612	0.8279
2.3287	26.2314	1600	1.8747	0.4748	0.9423
1.9794	29.5124	1800	1.7505	0.6135	0.9685

Framework versions

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