mms-1b-all-lg-CV-Fleurs-5hrs-v2

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: 0.4019
• Wer: 0.3812
• Cer: 0.0755

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
10.223	1.0	163	4.6201	1.0	0.9845
4.1813	2.0	326	3.8650	1.0	0.9602
3.6951	3.0	489	3.4980	1.0	0.8576
1.9773	4.0	652	0.3857	0.4087	0.0835
0.3493	5.0	815	0.3024	0.3692	0.0759
0.3096	6.0	978	0.2842	0.3626	0.0742
0.2962	7.0	1141	0.2740	0.3553	0.0730
0.2934	8.0	1304	0.2690	0.3547	0.0720
0.2843	9.0	1467	0.2658	0.3523	0.0720
0.2816	10.0	1630	0.2632	0.3506	0.0716
0.2733	11.0	1793	0.2610	0.3512	0.0715
0.2737	12.0	1956	0.2603	0.3501	0.0714
0.2723	13.0	2119	0.2611	0.3501	0.0712
0.2765	14.0	2282	0.2653	0.3515	0.0710
0.2819	15.0	2445	0.2796	0.3543	0.0713
0.2989	16.0	2608	0.2987	0.3565	0.0715
0.3175	17.0	2771	0.3202	0.3643	0.0731
0.3415	18.0	2934	0.3152	0.3659	0.0737
0.363	19.0	3097	0.3565	0.3662	0.0738
0.398	20.0	3260	0.4019	0.3812	0.0755

Framework versions

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