mms-1b-ckt-staged-bible-radio-to-chuklang-stage2

This model was trained from scratch.

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: 16
• eval_batch_size: 16
• seed: 42
• 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: 25
• num_epochs: 36
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.2669	1.1905	50	0.9174	0.8857	0.2330
1.0527	2.3810	100	0.8248	0.8333	0.2129
0.9774	3.5714	150	0.7784	0.8397	0.2076
0.9478	4.7619	200	0.7478	0.7794	0.1899
0.8638	5.9524	250	0.7369	0.7698	0.1871
0.8391	7.1429	300	0.6983	0.7714	0.1800
0.8098	8.3333	350	0.7036	0.7524	0.1767
0.7564	9.5238	400	0.7064	0.7302	0.1769
0.7321	10.7143	450	0.6941	0.7508	0.1743
0.688	11.9048	500	0.6999	0.7365	0.1754
0.694	13.0952	550	0.6810	0.7222	0.1719
0.6517	14.2857	600	0.6653	0.7127	0.1694
0.6306	15.4762	650	0.6734	0.7270	0.1696
0.5934	16.6667	700	0.6706	0.7302	0.1676
0.573	17.8571	750	0.6728	0.7127	0.1646
0.5911	19.0476	800	0.6692	0.7286	0.1681
0.5578	20.2381	850	0.6581	0.7159	0.1677
0.5402	21.4286	900	0.6643	0.7048	0.1661

Framework versions

• Transformers 4.57.1
• Pytorch 2.11.0+cu128
• Datasets 4.8.5
• Tokenizers 0.22.2