mms-1b-all-bemgen-combined-sd

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: 0.2536
• Wer: 0.4060

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: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.2531	0.5076	100	3.6265	0.9999
3.4512	1.0152	200	3.2377	1.0629
2.9585	1.5228	300	1.0987	0.8908
0.581	2.0305	400	0.2908	0.4380
0.467	2.5381	500	0.2799	0.4300
0.458	3.0457	600	0.2690	0.4104
0.4418	3.5533	700	0.2687	0.4162
0.4236	4.0609	800	0.2670	0.4072
0.4168	4.5685	900	0.2615	0.3989
0.4134	5.0761	1000	0.2588	0.3984
0.402	5.5838	1100	0.2583	0.3943
0.392	6.0914	1200	0.2582	0.3978
0.3907	6.5990	1300	0.2582	0.3967
0.3863	7.1066	1400	0.2571	0.3968
0.3873	7.6142	1500	0.2549	0.3978
0.3787	8.1218	1600	0.2536	0.4059
0.3785	8.6294	1700	0.2523	0.3902
0.3706	9.1371	1800	0.2539	0.3938
0.3673	9.6447	1900	0.2528	0.3907
0.3684	10.1523	2000	0.2539	0.3937
0.3638	10.6599	2100	0.2521	0.3833
0.3667	11.1675	2200	0.2512	0.3851
0.3706	11.6751	2300	0.2519	0.3881
0.3547	12.1827	2400	0.2532	0.3895
0.3465	12.6904	2500	0.2506	0.3877
0.3596	13.1980	2600	0.2529	0.3861
0.3497	13.7056	2700	0.2478	0.3855
0.3425	14.2132	2800	0.2486	0.3902
0.35	14.7208	2900	0.2572	0.3914
0.3547	15.2284	3000	0.2525	0.3894
0.3431	15.7360	3100	0.2500	0.3905

Framework versions

• Transformers 4.52.4
• Pytorch 2.9.0+cu128
• Datasets 4.4.1
• Tokenizers 0.21.4