mms-1b-bigcgen-combined-25hrs-model

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

• Loss: inf
• Wer: 0.5156

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: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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
• training_steps: 2500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
14.5485	0.0611	100	inf	1.0039
6.1502	0.1222	200	inf	1.0675
5.1685	0.1833	300	inf	1.0053
2.0876	0.2443	400	inf	0.5857
1.7116	0.3054	500	inf	0.5759
1.6505	0.3665	600	inf	0.5579
1.6573	0.4276	700	inf	0.5471
1.4679	0.4887	800	inf	0.5528
1.4955	0.5498	900	inf	0.5369
1.664	0.6109	1000	inf	0.5328
1.61	0.6720	1100	inf	0.5335
1.6414	0.7330	1200	inf	0.5293
1.6321	0.7941	1300	inf	0.5271
1.4686	0.8552	1400	inf	0.5297
1.5073	0.9163	1500	inf	0.5326
1.6164	0.9774	1600	inf	0.5235
1.577	1.0385	1700	inf	0.5238
1.383	1.0996	1800	inf	0.5217
1.4391	1.1607	1900	inf	0.5292
1.5327	1.2217	2000	inf	0.5255
1.3653	1.2828	2100	inf	0.5195
1.4901	1.3439	2200	inf	0.5187
1.4263	1.4050	2300	inf	0.5169
1.4603	1.4661	2400	inf	0.5179
1.4802	1.5272	2500	inf	0.5155

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0