mms-1B_all_nchlt_speech_corpus_Fleurs_CV_AFRIKAANS_57hr_v1

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

• Loss: 0.2848
• Wer: 0.3509
• Cer: 0.0823

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: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• 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: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8707	1.0	4001	0.4344	0.2868	0.0838
0.182	2.0	8002	0.4192	0.2555	0.0784
0.1582	3.0	12003	0.4040	0.2475	0.0764
0.1455	4.0	16004	0.3913	0.2407	0.0761
0.135	5.0	20005	0.3936	0.2321	0.0739
0.1274	6.0	24006	0.3862	0.2277	0.0734
0.1218	7.0	28007	0.3754	0.2284	0.0716
0.117	8.0	32008	0.3853	0.2331	0.0718
0.1112	9.0	36009	0.4186	0.2443	0.0751
0.1087	10.0	40010	0.3812	0.2282	0.0716
0.1052	11.0	44011	0.3810	0.2295	0.0735
0.101	12.0	48012	0.3622	0.2324	0.0714
0.0974	13.0	52013	0.4002	0.2391	0.0758
0.0965	14.0	56014	0.3738	0.2238	0.0703
0.0931	15.0	60015	0.3906	0.2347	0.0732
0.0909	16.0	64016	0.4118	0.2386	0.0750
0.088	17.0	68017	0.3725	0.2321	0.0726
0.0864	18.0	72018	0.3660	0.2357	0.0720
0.0858	19.0	76019	0.3887	0.2318	0.0717
0.083	20.0	80020	0.4284	0.2420	0.0742
0.083	21.0	84021	0.3996	0.2305	0.0713
0.0809	22.0	88022	0.4152	0.2321	0.0721
0.0792	23.0	92023	0.3987	0.2305	0.0717
0.0782	24.0	96024	0.3928	0.2305	0.0718

Framework versions

• Transformers 4.47.0
• Pytorch 2.1.0+cu118
• Datasets 3.2.0
• Tokenizers 0.21.0