mms-1b-all-lg-CV-Fleurs-20hrs-v6

This model is a fine-tuned version of facebook/mms-1b-all on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.9799
• Wer: 0.7317
• Cer: 0.1243

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: 5e-05
• 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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
6.0989	1.0	665	3.9815	1.0	0.9257
1.2245	2.0	1330	0.2702	0.3583	0.0726
0.317	3.0	1995	0.2566	0.3500	0.0711
0.3015	4.0	2660	0.2512	0.3465	0.0698
0.2971	5.0	3325	0.2487	0.3443	0.0691
0.294	6.0	3990	0.2478	0.3438	0.0690
0.293	7.0	4655	0.2490	0.3419	0.0688
0.2929	8.0	5320	0.2498	0.3429	0.0693
0.3039	9.0	5985	0.2670	0.3464	0.0699
0.3544	10.0	6650	0.3525	0.3885	0.0761
0.55	11.0	7315	0.7178	0.4961	0.0937
1.1365	12.0	7980	1.1812	0.8922	0.2352
1.0735	13.0	8645	0.9729	0.6441	0.1298
1.0835	14.0	9310	1.0791	0.5851	0.1111
1.1183	15.0	9975	1.0701	0.5538	0.1009
1.0588	16.0	10640	0.9951	0.7190	0.1214
1.0276	17.0	11305	0.9799	0.7317	0.1243

Framework versions

• Transformers 4.48.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0