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

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: 3.9548
• Cer: 0.9538
• Wer: 1.0010

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_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
1.8351	3.2810	200	1.3899	0.1624	0.5561
1.5048	6.5620	400	1.2824	0.1444	0.4732
1.5952	9.8430	600	1.2806	0.1387	0.4533
0.7156	13.1157	800	1.5784	0.1426	0.4701
0.7436	16.3967	1000	1.3450	0.1473	0.4680
2.1968	19.6777	1200	1.6456	0.5245	0.9318
2.8915	22.9587	1400	2.7871	0.7356	0.9874
3.9703	26.2314	1600	3.8258	0.8417	1.0010
4.0352	29.5124	1800	3.9548	0.9538	1.0010

Framework versions

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