mms-1b-bigcgen-combined-10hrs-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: inf
• Wer: 0.5167

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
• training_steps: 2500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
14.705	0.1528	100	inf	1.0004
6.0879	0.3056	200	inf	1.0203
5.4042	0.4584	300	inf	1.0160
2.8156	0.6112	400	inf	0.6058
1.9111	0.7639	500	inf	0.5724
1.82	0.9167	600	inf	0.5643
1.6294	1.0688	700	inf	0.5685
1.6856	1.2215	800	inf	0.5530
1.6363	1.3743	900	inf	0.5501
1.5114	1.5271	1000	inf	0.5417
1.5417	1.6799	1100	inf	0.5358
1.6518	1.8327	1200	inf	0.5337
1.4795	1.9855	1300	inf	0.5292
1.5822	2.1375	1400	inf	0.5278
1.4938	2.2903	1500	inf	0.5194
1.5701	2.4431	1600	inf	0.5353
1.47	2.5959	1700	inf	0.5183
1.4109	2.7487	1800	inf	0.5342
1.3993	2.9015	1900	inf	0.5175
1.4848	3.0535	2000	inf	0.5177
1.4331	3.2063	2100	inf	0.5286
1.4392	3.3591	2200	inf	0.5213
1.3324	3.5118	2300	inf	0.5161
1.501	3.6646	2400	inf	0.5160
1.3526	3.8174	2500	inf	0.5163

Framework versions

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