asr-africa/mms-1b-all-lin-Fleurs_AMMI_AFRIVOICE_LRSC-5hrs-v2

mms-1b-all-lin-Fleurs_AMMI_AFRIVOICE_LRSC-10hrs-v1

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: 0.3102
• Wer: 0.2115
• Cer: 0.0625

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: 0.001
• 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
5.5986	1.0	192	0.4551	0.3955	0.1115
0.3962	2.0	384	0.3596	0.2570	0.0738
0.343	3.0	576	0.3257	0.2582	0.0741
0.3239	4.0	768	0.3222	0.2852	0.0776
0.31	5.0	960	0.3264	0.2473	0.0711
0.2946	6.0	1152	0.3234	0.2223	0.0669
0.2908	7.0	1344	0.3038	0.2252	0.0650
0.2717	8.0	1536	0.3142	0.2399	0.0691
0.2652	9.0	1728	0.3254	0.2348	0.0677
0.2555	10.0	1920	0.3079	0.2196	0.0648
0.2527	11.0	2112	0.3156	0.2165	0.0646
0.2467	12.0	2304	0.3040	0.2250	0.0652
0.2318	13.0	2496	0.3031	0.2109	0.0626
0.2259	14.0	2688	0.3067	0.2155	0.0637
0.219	15.0	2880	0.3094	0.2066	0.0617
0.2174	16.0	3072	0.3107	0.2079	0.0620
0.2115	17.0	3264	0.3057	0.2140	0.0631
0.2056	18.0	3456	0.3192	0.2104	0.0625
0.2029	19.0	3648	0.3042	0.2116	0.0626
0.196	20.0	3840	0.3097	0.2165	0.0632
0.1874	21.0	4032	0.3098	0.2133	0.0663
0.1829	22.0	4224	0.3108	0.2090	0.0627
0.1823	23.0	4416	0.3316	0.2292	0.0672
0.1756	24.0	4608	0.3194	0.2166	0.0639
0.1725	25.0	4800	0.3102	0.2115	0.0625

Framework versions

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