mms-1b-all-bigc-combined-20hrs-52

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

• Loss: 0.8685
• Wer: 0.5988

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: 52
• 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: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.6068	0.1623	100	2.0099	0.8922
0.8321	0.3247	200	1.4605	0.7303
0.7144	0.4870	300	1.4735	0.7223
0.6934	0.6494	400	1.4167	0.7039
0.7245	0.8117	500	1.3788	0.6966
0.6651	0.9740	600	1.3661	0.6978
0.6636	1.1364	700	1.3663	0.6941
0.6575	1.2987	800	1.3906	0.6858
0.6883	1.4610	900	1.2550	0.6731
0.6542	1.6234	1000	1.2789	0.6732
0.6892	1.7857	1100	1.2231	0.6663
0.5936	1.9481	1200	1.2362	0.6699
0.6526	2.1104	1300	1.0530	0.6570
0.6443	2.2727	1400	1.0636	0.6466
0.6239	2.4351	1500	1.1153	0.6553
0.6382	2.5974	1600	1.0441	0.6675
0.6003	2.7597	1700	1.0318	0.6427
0.5535	2.9221	1800	0.9456	0.6432
0.5992	3.0844	1900	0.9456	0.6335
0.5649	3.2468	2000	0.8685	0.5989
0.5744	3.4091	2100	0.9052	0.6203
0.6067	3.5714	2200	0.8639	0.6116
0.5649	3.7338	2300	1.0003	0.6330
0.6078	3.8961	2400	0.8985	0.6196
0.5816	4.0584	2500	0.8801	0.6105

Framework versions

• Transformers 4.53.0.dev0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.0