ssc-top-mms-model-mix-adapt-max

This model is a fine-tuned version of facebook/mms-1b-all on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7289
• Cer: 0.1445
• Wer: 0.5792

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: 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: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.3243	0.3524	200	0.8187	0.1612	0.6843
1.2349	0.7048	400	0.7650	0.1595	0.6707
1.2417	1.0564	600	0.7351	0.1513	0.6322
1.2134	1.4088	800	0.7760	0.1521	0.6384
1.1341	1.7612	1000	0.7522	0.1478	0.5995
1.1513	2.1128	1200	0.7333	0.1489	0.6002
1.0074	2.4652	1400	0.7404	0.1495	0.6197
1.0128	2.8176	1600	0.7439	0.1515	0.6232
1.0346	3.1692	1800	0.7357	0.1478	0.6088
1.005	3.5216	2000	0.7420	0.1483	0.6065
0.9878	3.8740	2200	0.7508	0.1468	0.6041
0.9541	4.2256	2400	0.7342	0.1470	0.6014
0.9563	4.5780	2600	0.7314	0.1449	0.5812
0.9478	4.9304	2800	0.7289	0.1445	0.5792

Framework versions

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