mms-1b-all-swagen-combined-5hrs-42

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

• Loss: 0.3301
• Wer: 0.2354

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
7.2606	0.4785	100	3.1592	1.0085
2.7521	0.9569	200	2.3371	0.9931
2.0221	1.4354	300	1.5196	0.8476
1.152	1.9139	400	0.7881	0.4487
0.7455	2.3923	500	0.6427	0.3592
0.6182	2.8708	600	0.5445	0.3369
0.5513	3.3493	700	0.4796	0.3189
0.4933	3.8278	800	0.4499	0.3112
0.441	4.3062	900	0.4219	0.2904
0.4248	4.7847	1000	0.4193	0.2819
0.4207	5.2632	1100	0.3872	0.2785
0.3842	5.7416	1200	0.3788	0.2687
0.3717	6.2201	1300	0.3735	0.2650
0.36	6.6986	1400	0.3771	0.2620
0.3503	7.1770	1500	0.3682	0.2526
0.3373	7.6555	1600	0.3540	0.2519
0.3404	8.1340	1700	0.3481	0.2452
0.3347	8.6124	1800	0.3396	0.2484
0.327	9.0909	1900	0.3536	0.2427
0.3035	9.5694	2000	0.3567	0.2420
0.3232	10.0478	2100	0.3385	0.2439
0.3051	10.5263	2200	0.3420	0.2398
0.3093	11.0048	2300	0.3330	0.2331
0.3013	11.4833	2400	0.3301	0.2354
0.2975	11.9617	2500	0.3385	0.2336
0.2741	12.4402	2600	0.3271	0.2305
0.2794	12.9187	2700	0.3289	0.2279
0.282	13.3971	2800	0.3331	0.2284
0.2711	13.8756	2900	0.3350	0.2267

Framework versions

• Transformers 4.53.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.0