mms-1b-swagen-female-model

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

• Loss: 0.2065
• Wer: 0.1900

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
• 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
8.5907	0.1176	100	4.5590	1.0
4.3006	0.2353	200	4.2573	1.0
4.1906	0.3529	300	3.8436	1.0
1.1198	0.4706	400	0.2569	0.1996
0.3012	0.5882	500	0.2466	0.1942
0.2919	0.7059	600	0.2501	0.1990
0.267	0.8235	700	0.2353	0.1894
0.2666	0.9412	800	0.2280	0.1904
0.2396	1.0588	900	0.2238	0.1892
0.258	1.1765	1000	0.2202	0.1898
0.2192	1.2941	1100	0.2196	0.1938
0.2353	1.4118	1200	0.2169	0.1875
0.2398	1.5294	1300	0.2164	0.1896
0.2419	1.6471	1400	0.2146	0.1913
0.2582	1.7647	1500	0.2122	0.1905
0.2417	1.8824	1600	0.2105	0.1861
0.2375	2.0	1700	0.2091	0.1892
0.2353	2.1176	1800	0.2087	0.1882
0.23	2.2353	1900	0.2091	0.1904
0.2378	2.3529	2000	0.2067	0.1898
0.2343	2.4706	2100	0.2061	0.1890
0.2083	2.5882	2200	0.2084	0.1884
0.2058	2.7059	2300	0.2085	0.1882
0.2197	2.8235	2400	0.2062	0.1904
0.2317	2.9412	2500	0.2065	0.1896

Framework versions

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