mms-1b-nyagen-combined-model

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

• Loss: 0.1727
• Wer: 0.2465

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
• 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
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.9978	0.1364	100	0.6384	0.5015
0.482	0.2729	200	0.2777	0.3713
0.3907	0.4093	300	0.2484	0.3481
0.3782	0.5457	400	0.2290	0.3232
0.3316	0.6821	500	0.2222	0.3148
0.3158	0.8186	600	0.2127	0.3042
0.3199	0.9550	700	0.2106	0.2932
0.3223	1.0914	800	0.2013	0.2826
0.3075	1.2278	900	0.1975	0.2709
0.3015	1.3643	1000	0.1942	0.2762
0.3049	1.5007	1100	0.1895	0.2729
0.3029	1.6371	1200	0.1888	0.2718
0.2626	1.7735	1300	0.1866	0.2683
0.2803	1.9100	1400	0.1830	0.2615
0.2725	2.0464	1500	0.1814	0.2626
0.2732	2.1828	1600	0.1783	0.2641
0.249	2.3192	1700	0.1828	0.2560
0.2423	2.4557	1800	0.1762	0.2480
0.2668	2.5921	1900	0.1732	0.2458
0.2653	2.7285	2000	0.1727	0.2460
0.2614	2.8649	2100	0.1749	0.2533
0.2474	3.0014	2200	0.1733	0.2438
0.2317	3.1378	2300	0.1767	0.2447

Framework versions

• Transformers 4.48.0.dev0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0