mms-1b-all-bemgen-combined-sd-1e-0

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: 1.6674
• Wer: 0.4200

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: 0.000275
• 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
8.2742	0.5076	100	4.9931	0.9998
4.7522	1.0152	200	4.5563	1.0079
4.3117	1.5228	300	3.5193	0.9999
2.2552	2.0305	400	1.7639	0.4427
1.897	2.5381	500	1.7190	0.4245
1.8635	3.0457	600	1.6986	0.4155
1.8428	3.5533	700	1.6909	0.4437
1.8221	4.0609	800	1.6860	0.4046
1.8201	4.5685	900	1.6846	0.4118
1.8056	5.0761	1000	1.6819	0.4067
1.8023	5.5838	1100	1.6789	0.4234
1.7851	6.0914	1200	1.6751	0.4505
1.7847	6.5990	1300	1.6757	0.4221
1.7841	7.1066	1400	1.6742	0.4082
1.7895	7.6142	1500	1.6732	0.4238
1.7754	8.1218	1600	1.6718	0.4478
1.7766	8.6294	1700	1.6739	0.4180
1.7677	9.1371	1800	1.6700	0.4344
1.7688	9.6447	1900	1.6686	0.4184
1.761	10.1523	2000	1.6682	0.4188
1.7637	10.6599	2100	1.6680	0.4521
1.7631	11.1675	2200	1.6697	0.4530
1.7744	11.6751	2300	1.6630	0.4169
1.7518	12.1827	2400	1.6659	0.4193
1.7476	12.6904	2500	1.6618	0.4385
1.7591	13.1980	2600	1.6645	0.4384
1.7486	13.7056	2700	1.6612	0.4367
1.7424	14.2132	2800	1.6612	0.4379
1.7528	14.7208	2900	1.6627	0.4076
1.753	15.2284	3000	1.6610	0.4288
1.7472	15.7360	3100	1.6635	0.4311
1.7424	16.2437	3200	1.6627	0.4025
1.7416	16.7513	3300	1.6614	0.4071
1.7367	17.2589	3400	1.6594	0.4278
1.745	17.7665	3500	1.6609	0.4300
1.7374	18.2741	3600	1.6648	0.4279
1.7398	18.7817	3700	1.6606	0.4150
1.7279	19.2893	3800	1.6586	0.4459
1.74	19.7970	3900	1.6593	0.3867
1.7287	20.3046	4000	1.6605	0.4226
1.7347	20.8122	4100	1.6586	0.4363
1.727	21.3198	4200	1.6576	0.3938
1.7314	21.8274	4300	1.6562	0.4252
1.7209	22.3350	4400	1.6572	0.3905
1.7282	22.8426	4500	1.6576	0.4304
1.7323	23.3503	4600	1.6568	0.4132
1.7159	23.8579	4700	1.6555	0.4313
1.7188	24.3655	4800	1.6574	0.4013
1.7301	24.8731	4900	1.6568	0.4254
1.7197	25.3807	5000	1.6560	0.4436
1.7211	25.8883	5100	1.6560	0.4274

Framework versions

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