mms-1b-all-bemgen-combined-vanilla

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

• Loss: 0.2360
• Wer: 0.4322

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: 4
• 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: 5.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.7513	0.5076	100	5.6808	1.0
4.8332	1.0152	200	5.0254	1.0181
4.3374	1.5228	300	4.1060	1.0146
3.5458	2.0305	400	3.0970	0.9987
1.1532	2.5381	500	0.2786	0.4790
0.4689	3.0457	600	0.2472	0.4479
0.4451	3.5533	700	0.2422	0.4390
0.4178	4.0609	800	0.2360	0.4322
0.4117	4.5685	900	0.2305	0.4170

Framework versions

• Transformers 4.53.0
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.21.4