mms-1b-all-bem-natbed-n-model

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

• Loss: 0.5309
• Wer: 0.4631

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.3623	0.2809	100	0.9287	0.7283
0.9213	0.5618	200	0.6511	0.5931
0.7224	0.8427	300	0.6387	0.5434
0.7132	1.1236	400	0.6140	0.5213
0.7195	1.4045	500	0.6097	0.5146
0.7054	1.6854	600	0.6145	0.5126
0.7417	1.9663	700	0.6062	0.5257
0.7029	2.2472	800	0.6022	0.4947
0.6845	2.5281	900	0.5886	0.5023
0.663	2.8090	1000	0.5915	0.4926
0.7129	3.0899	1100	0.5833	0.4920
0.6735	3.3708	1200	0.5877	0.4832
0.672	3.6517	1300	0.5863	0.5151
0.6494	3.9326	1400	0.5795	0.4844
0.7049	4.2135	1500	0.5724	0.4716
0.5898	4.4944	1600	0.5640	0.4762
0.6581	4.7753	1700	0.5582	0.4724
0.6262	5.0562	1800	0.5447	0.4751
0.6179	5.3371	1900	0.5497	0.4656
0.5896	5.6180	2000	0.5444	0.4779
0.6438	5.8989	2100	0.5399	0.4700
0.6086	6.1798	2200	0.5520	0.4598
0.6226	6.4607	2300	0.5386	0.4797
0.6148	6.7416	2400	0.5574	0.4680
0.5838	7.0225	2500	0.5497	0.4639
0.5407	7.3034	2600	0.5377	0.4631
0.6186	7.5843	2700	0.5404	0.4715
0.5922	7.8652	2800	0.5381	0.4609
0.5799	8.1461	2900	0.5312	0.4620
0.5914	8.4270	3000	0.5309	0.4631
0.6194	8.7079	3100	0.5317	0.4678
0.5851	8.9888	3200	0.5389	0.4575
0.5764	9.2697	3300	0.5579	0.4550

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0