mms-1b-all-bemgen-combined-gdro-new

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

• Loss: 17.2593
• Wer: 0.5187

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: 4
• 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: 100
• num_epochs: 5.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
642.4339	0.5076	100	196.5480	1.0
187.0374	1.0152	200	40.3816	0.7633
113.298	1.5228	300	20.5291	0.5622
96.5571	2.0305	400	18.9103	0.5312
94.3258	2.5381	500	18.2364	0.5210
89.2214	3.0457	600	17.7855	0.5297
89.8372	3.5533	700	17.7325	0.5394
88.9666	4.0609	800	17.2595	0.5187
88.3331	4.5685	900	17.2089	0.5121

Framework versions

• Transformers 4.52.4
• Pytorch 2.9.0+cu128
• Datasets 4.4.1
• Tokenizers 0.21.4