mms-1b-swagen-combined-10hrs-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: 0.2348
• Wer: 0.1949

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: 2500.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
13.9483	0.1195	100	3.5419	1.0006
6.4016	0.2391	200	2.8641	0.9930
2.3599	0.3586	300	0.2764	0.2088
0.5781	0.4782	400	0.2576	0.1993
0.5212	0.5977	500	0.2503	0.1969
0.5119	0.7173	600	0.2414	0.1951
0.5046	0.8368	700	0.2435	0.1929
0.5252	0.9564	800	0.2351	0.1937
0.4838	1.0753	900	0.2330	0.1941
0.4479	1.1949	1000	0.2352	0.2010
0.4541	1.3144	1100	0.2384	0.1946
0.4366	1.4340	1200	0.2382	0.1952
0.4919	1.5535	1300	0.2348	0.1947

Framework versions

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