metadata

library_name: transformers
license: cc-by-nc-4.0
base_model: facebook/mms-1b-all
tags:
  - automatic-speech-recognition
  - natbed
  - mms
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: mms-1b-all-bem-natbed-nn-model
    results: []

mms-1b-all-bem-natbed-nn-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: 0.5884
Wer: 0.5333

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
7.9244	0.2257	100	1.3514	1.0238
1.0236	0.4515	200	0.8355	0.6595
0.8005	0.6772	300	0.7837	0.6141
0.8968	0.9029	400	0.7809	0.6043
0.8909	1.1287	500	0.7147	0.5953
0.7983	1.3544	600	0.6990	0.5931
0.8563	1.5801	700	0.6805	0.5965
0.7094	1.8059	800	0.6849	0.5808
0.7499	2.0316	900	0.6457	0.5934
0.7722	2.2573	1000	0.6565	0.5875
0.7099	2.4831	1100	0.6419	0.5596
0.7416	2.7088	1200	0.6195	0.5611
0.6385	2.9345	1300	0.6228	0.5647
0.6436	3.1603	1400	0.6184	0.5510
0.6795	3.3860	1500	0.6157	0.5533
0.7027	3.6117	1600	0.6343	0.5426
0.6585	3.8375	1700	0.6057	0.5428
0.6351	4.0632	1800	0.6017	0.5430
0.6528	4.2889	1900	0.6099	0.5340
0.6603	4.5147	2000	0.6218	0.5335
0.6676	4.7404	2100	0.5977	0.5323
0.6304	4.9661	2200	0.5884	0.5333
0.5976	5.1919	2300	0.5956	0.5228
0.6564	5.4176	2400	0.5957	0.5302
0.6717	5.6433	2500	0.5767	0.5183
0.6091	5.8691	2600	0.5921	0.5273
0.6168	6.0948	2700	0.5894	0.5275
0.6495	6.3205	2800	0.6036	0.5197

Framework versions

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