mms-1b-bigcgen-male-15hrs-model

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

• Loss: 0.4756
• Wer: 0.4703

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use 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: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
14.7965	0.1033	100	3.5227	1.0007
6.2688	0.2066	200	2.7700	1.0325
3.6179	0.3099	300	0.7532	0.6069
1.7779	0.4132	400	0.6508	0.5821
1.5595	0.5165	500	0.6249	0.5578
1.5884	0.6198	600	0.6142	0.5283
1.5532	0.7231	700	0.5929	0.5172
1.4021	0.8264	800	0.5996	0.5182
1.507	0.9298	900	0.5824	0.5121
1.5374	1.0331	1000	0.5615	0.5061
1.4139	1.1364	1100	0.5456	0.5066
1.4472	1.2397	1200	0.5177	0.4874
1.2958	1.3430	1300	0.5022	0.4871
1.3292	1.4463	1400	0.4984	0.4871
1.2062	1.5496	1500	0.4886	0.4799
1.1623	1.6529	1600	0.4811	0.4823
1.2759	1.7562	1700	0.4735	0.4677
1.1852	1.8595	1800	0.4986	0.4669
1.0712	1.9628	1900	0.5045	0.4845
1.2023	2.0661	2000	0.4755	0.4782
1.2275	2.1694	2100	0.4756	0.4705

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0