wav2vec2-large-mms-1b-dzo_M5

This model is a fine-tuned version of facebook/mms-1b-all on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2842
• Wer: 0.3115

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: 0.004
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
7.869	1.0	32	5.7328	1.0578
3.1799	2.0	64	2.5788	0.9945
1.1289	3.0	96	0.8093	0.5926
0.743	4.0	128	0.6091	0.5035
0.6609	5.0	160	0.5270	0.4642
0.6302	6.0	192	0.4892	0.4301
0.6035	7.0	224	0.4419	0.4013
0.5367	8.0	256	0.4096	0.3844
0.5291	9.0	288	0.4008	0.3883
0.4851	10.0	320	0.3602	0.3634
0.4994	11.0	352	0.3457	0.3532
0.4523	12.0	384	0.3314	0.3514
0.4884	13.0	416	0.3278	0.3414
0.4278	14.0	448	0.3323	0.3471
0.4011	14.5397	465	0.2842	0.3115

Framework versions

• Transformers 4.49.0.dev0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0