mms-1B_all_NCHLT_speech_corpus_Fleurs_Zulu_63hr_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: 0.1187
• Wer: 0.2365
• Cer: 0.0422

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
1.6864	1.0	2755	0.2577	0.3780	0.0648
0.213	2.0	5510	0.2430	0.3445	0.0589
0.1965	3.0	8265	0.2381	0.3316	0.0589
0.1845	4.0	11020	0.2371	0.3268	0.0580
0.176	5.0	13775	0.2439	0.3283	0.0584
0.1681	6.0	16530	0.2407	0.3307	0.0580
0.1635	7.0	19285	0.2488	0.3414	0.0596
0.1582	8.0	22040	0.2443	0.3261	0.0584
0.1515	9.0	24795	0.2417	0.3228	0.0577
0.149	10.0	27550	0.2382	0.3169	0.0559
0.1448	11.0	30305	0.2442	0.3276	0.0573
0.1419	12.0	33060	0.2387	0.3206	0.0574
0.1392	13.0	35815	0.2402	0.3281	0.0569
0.1354	14.0	38570	0.2474	0.3266	0.0579
0.133	15.0	41325	0.2399	0.3257	0.0569
0.1295	16.0	44080	0.2336	0.3099	0.0546
0.1278	17.0	46835	0.2396	0.3174	0.0569
0.1254	18.0	49590	0.2382	0.3150	0.0558
0.1228	19.0	52345	0.2385	0.3152	0.0554
0.1202	20.0	55100	0.2359	0.3095	0.0544
0.1181	21.0	57855	0.2444	0.3143	0.0561
0.1168	22.0	60610	0.2384	0.3171	0.0553
0.1136	23.0	63365	0.2349	0.3093	0.0548
0.1124	24.0	66120	0.2429	0.3178	0.0562
0.1107	25.0	68875	0.2340	0.3106	0.0549
0.1089	26.0	71630	0.2375	0.3115	0.0555

Framework versions

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