mms-1b-bemgen-100f50m-model

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: 0.2051
• Wer: 0.3518

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
13.3814	0.1374	100	0.6632	0.7196
1.0155	0.2747	200	0.3021	0.4662
0.8042	0.4121	300	0.2792	0.4686
0.7806	0.5495	400	0.2685	0.4386
0.7457	0.6868	500	0.2673	0.4402
0.7131	0.8242	600	0.2641	0.4238
0.7382	0.9615	700	0.2537	0.4253
0.6811	1.0989	800	0.2501	0.4225
0.6988	1.2363	900	0.2492	0.4222
0.6857	1.3736	1000	0.2475	0.4058
0.6711	1.5110	1100	0.2455	0.4072
0.6802	1.6484	1200	0.2403	0.4068
0.6962	1.7857	1300	0.2387	0.4112
0.6656	1.9231	1400	0.2400	0.4154
0.6643	2.0604	1500	0.2355	0.4047
0.6502	2.1978	1600	0.2405	0.4032
0.6309	2.3352	1700	0.2428	0.4530
0.6572	2.4725	1800	0.2320	0.3956
0.6026	2.6099	1900	0.2353	0.3874
0.6578	2.7473	2000	0.2298	0.4008
0.6493	2.8846	2100	0.2284	0.3864
0.609	3.0220	2200	0.2280	0.3858
0.611	3.1593	2300	0.2259	0.3883
0.5993	3.2967	2400	0.2237	0.3832
0.5881	3.4341	2500	0.2230	0.3792
0.6212	3.5714	2600	0.2214	0.3810
0.5731	3.7088	2700	0.2220	0.3689
0.6249	3.8462	2800	0.2204	0.3695
0.6344	3.9835	2900	0.2187	0.3677
0.5931	4.1209	3000	0.2188	0.3745
0.5805	4.2582	3100	0.2160	0.3761
0.606	4.3956	3200	0.2159	0.3803
0.6137	4.5330	3300	0.2164	0.3894
0.5491	4.6703	3400	0.2146	0.3700
0.5659	4.8077	3500	0.2161	0.3596
0.5802	4.9451	3600	0.2135	0.3657
0.5921	5.0824	3700	0.2128	0.3582
0.5528	5.2198	3800	0.2107	0.3625
0.5584	5.3571	3900	0.2152	0.3878
0.5712	5.4945	4000	0.2129	0.3556
0.5863	5.6319	4100	0.2103	0.3536
0.5404	5.7692	4200	0.2110	0.3559
0.5715	5.9066	4300	0.2116	0.3541
0.5688	6.0440	4400	0.2137	0.3640
0.5249	6.1813	4500	0.2098	0.3647
0.5489	6.3187	4600	0.2098	0.3515
0.5325	6.4560	4700	0.2080	0.3563
0.5517	6.5934	4800	0.2105	0.3630
0.5392	6.7308	4900	0.2102	0.3651
0.5513	6.8681	5000	0.2052	0.3440
0.5526	7.0055	5100	0.2062	0.3403
0.5306	7.1429	5200	0.2069	0.3659
0.5021	7.2802	5300	0.2065	0.3520
0.5706	7.4176	5400	0.2042	0.3515
0.5223	7.5549	5500	0.2029	0.3430
0.5516	7.6923	5600	0.2032	0.3436
0.5249	7.8297	5700	0.2068	0.3553
0.5028	7.9670	5800	0.2047	0.3511
0.4719	8.1044	5900	0.2051	0.3515

Framework versions

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