Kabardian-ASR-kaggle

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: 0.1748
• Wer: 0.3268

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: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.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: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6132	0.3040	200	0.4195	0.6846
0.4931	0.6079	400	0.3299	0.5413
0.3934	0.9119	600	0.2873	0.5070
0.3771	1.2158	800	0.2497	0.4569
0.3445	1.5198	1000	0.2516	0.4304
0.3724	1.8237	1200	0.2439	0.4079
0.3265	2.1277	1400	0.2083	0.4065
0.2983	2.4316	1600	0.2082	0.3693
0.3222	2.7356	1800	0.2073	0.3688
0.3461	3.0395	2000	0.1854	0.3438
0.2745	3.3435	2200	0.1813	0.3329
0.2867	3.6474	2400	0.1784	0.3258
0.2715	3.9514	2600	0.1748	0.3268

Framework versions

• Transformers 4.49.0.dev0
• Pytorch 2.5.1+cu124
• Datasets 3.3.0
• Tokenizers 0.21.0