mms-1b-all-bigc-combined-30hrs-52

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

• Loss: 1.0938
• Wer: 0.6645

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: 52
• 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: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.7325	0.1083	100	1.9183	0.9016
0.8428	0.2167	200	1.3853	0.7476
0.7825	0.3250	300	1.4461	0.7318
0.7408	0.4334	400	1.4676	0.7156
0.721	0.5417	500	1.3413	0.7042
0.7613	0.6501	600	1.3475	0.7026
0.6874	0.7584	700	1.3475	0.6943
0.6712	0.8667	800	1.2602	0.6920
0.7128	0.9751	900	1.2171	0.6878
0.6616	1.0834	1000	1.2130	0.6846
0.6838	1.1918	1100	1.2407	0.6797
0.6797	1.3001	1200	1.0939	0.6645
0.6873	1.4085	1300	1.0737	0.6705
0.6793	1.5168	1400	1.1411	0.6775
0.6821	1.6251	1500	1.0875	0.6674
0.6779	1.7335	1600	1.0898	0.6625

Framework versions

• Transformers 4.53.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.0