ssc-qxp-mms-model-mix-adapt-max

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: 0.1612
• Cer: 0.0900
• Wer: 0.5028

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: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.215	0.9975	200	0.1886	0.1071	0.5928
0.1815	1.9925	400	0.1718	0.1030	0.5625
0.1654	2.9875	600	0.1648	0.1021	0.5597
0.1564	3.9825	800	0.1603	0.1003	0.5478
0.146	4.9776	1000	0.1513	0.0957	0.5358
0.1375	5.9726	1200	0.1558	0.1001	0.5478
0.139	6.9676	1400	0.1593	0.0994	0.5368
0.1266	7.9626	1600	0.1467	0.0952	0.5322
0.1267	8.9576	1800	0.1552	0.0971	0.5432
0.1254	9.9526	2000	0.1536	0.0982	0.5377
0.1169	10.9476	2200	0.1497	0.0930	0.5193
0.115	11.9426	2400	0.1527	0.0965	0.5303
0.116	12.9377	2600	0.1505	0.0953	0.5239
0.1096	13.9327	2800	0.1560	0.0960	0.5303
0.1051	14.9277	3000	0.1599	0.0976	0.5340
0.1016	15.9227	3200	0.1541	0.0941	0.5156
0.0958	16.9177	3400	0.1592	0.0956	0.5303
0.0894	17.9127	3600	0.1573	0.0950	0.5211
0.0838	18.9077	3800	0.1567	0.0939	0.5257
0.0836	19.9027	4000	0.1631	0.0943	0.5175
0.0803	20.8978	4200	0.1612	0.0927	0.5156
0.0761	21.8928	4400	0.1526	0.0883	0.5037
0.0753	22.8878	4600	0.1589	0.0931	0.5184
0.0724	23.8828	4800	0.1597	0.0928	0.5129
0.0689	24.8778	5000	0.1622	0.0918	0.5101
0.0644	25.8728	5200	0.1620	0.0907	0.5055
0.066	26.8678	5400	0.1591	0.0888	0.5
0.0663	27.8628	5600	0.1590	0.0873	0.5
0.0659	28.8579	5800	0.1600	0.0894	0.5037
0.0615	29.8529	6000	0.1612	0.0900	0.5028

Framework versions

• Transformers 4.57.2
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.22.0