You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

facebook/mms-1b-all

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

  • Loss: 0.5796
  • Model Preparation Time: 0.0177
  • Wer: 0.4635
  • Cer: 0.1168

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
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Model Preparation Time Wer Cer
5.2767 1.0 154 0.6576 0.0177 0.5943 0.1359
0.7588 2.0 308 0.6085 0.0177 0.5476 0.1268
0.7192 3.0 462 0.5971 0.0177 0.5313 0.1248
0.7048 4.0 616 0.5944 0.0177 0.5432 0.1246
0.6911 5.0 770 0.5889 0.0177 0.5271 0.1226
0.679 6.0 924 0.5702 0.0177 0.5111 0.1205
0.6623 7.0 1078 0.5724 0.0177 0.5046 0.1201
0.65 8.0 1232 0.5605 0.0177 0.5054 0.1209
0.6421 9.0 1386 0.5515 0.0177 0.4993 0.1178
0.626 10.0 1540 0.5518 0.0177 0.4914 0.1166
0.5999 11.0 1694 0.5358 0.0177 0.4910 0.1161
0.5824 12.0 1848 0.5425 0.0177 0.5016 0.1226
0.5723 13.0 2002 0.5325 0.0177 0.5071 0.1195
0.5576 14.0 2156 0.5437 0.0177 0.4880 0.1160
0.5498 15.0 2310 0.5725 0.0177 0.5341 0.1466
0.5383 16.0 2464 0.5814 0.0177 0.4721 0.1141
0.5283 17.0 2618 0.5483 0.0177 0.4819 0.1170
0.5145 18.0 2772 0.5297 0.0177 0.4931 0.1237
0.4977 19.0 2926 0.5283 0.0177 0.4889 0.1208
0.4913 20.0 3080 0.5365 0.0177 0.4776 0.1236
0.4822 21.0 3234 0.5562 0.0177 0.4708 0.1149
0.4768 22.0 3388 0.5493 0.0177 0.4804 0.1160
0.4598 23.0 3542 0.5574 0.0177 0.4736 0.1165
0.4558 24.0 3696 0.5340 0.0177 0.4766 0.1195
0.4516 25.0 3850 0.5703 0.0177 0.4787 0.1143
0.44 26.0 4004 0.5329 0.0177 0.4662 0.1144
0.4322 27.0 4158 0.5790 0.0177 0.4738 0.1136
0.4264 28.0 4312 0.5581 0.0177 0.4729 0.1133
0.4193 29.0 4466 0.5655 0.0177 0.4642 0.1144
0.4115 30.0 4620 0.5521 0.0177 0.4657 0.1173
0.4046 31.0 4774 0.5370 0.0177 0.4626 0.1139
0.4037 32.0 4928 0.5517 0.0177 0.4753 0.1173
0.4016 33.0 5082 0.5733 0.0177 0.4566 0.1119
0.3928 34.0 5236 0.5542 0.0177 0.4715 0.1164
0.3827 35.0 5390 0.5504 0.0177 0.4587 0.1132
0.3828 36.0 5544 0.5541 0.0177 0.4587 0.1126
0.3788 37.0 5698 0.5548 0.0177 0.4551 0.1121
0.371 38.0 5852 0.5574 0.0177 0.4543 0.1131
0.3712 39.0 6006 0.5709 0.0177 0.4600 0.1114
0.3631 40.0 6160 0.5783 0.0177 0.4655 0.1174
0.3561 41.0 6314 0.5753 0.0177 0.4628 0.1151
0.3518 42.0 6468 0.5695 0.0177 0.4691 0.1188
0.3472 43.0 6622 0.5802 0.0177 0.4619 0.1119
0.3423 44.0 6776 0.5796 0.0177 0.4636 0.1146
0.3352 45.0 6930 0.5940 0.0177 0.4585 0.1145
0.3335 46.0 7084 0.5915 0.0177 0.4679 0.1189
0.3294 47.0 7238 0.5885 0.0177 0.4664 0.1165

Framework versions

  • Transformers 4.47.0.dev0
  • Pytorch 2.1.0+cu118
  • Datasets 3.1.0
  • Tokenizers 0.20.1
Downloads last month
0
Safetensors
Model size
965M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for asr-africa/mms-1B_all_BIG_C_Bemba_5hr_v1

Finetuned
(185)
this model

Evaluation results