w2vbert-waxal-wx

This model is a fine-tuned version of sulaimank/w2vbert-waxal-p2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4384
• Wer Ach: 0.3489
• Cer Ach: 0.1383
• Zindi Ach: 0.7564
• Wer Mas: 0.5010
• Cer Mas: 0.1094
• Zindi Mas: 0.6948
• Wer Nyn: 0.3660
• Cer Nyn: 0.0856
• Zindi Nyn: 0.7742
• Wer: 0.4078
• Cer: 0.1069
• Zindi: 0.7426
• Zindi Strip: 0.7700
• Zindi Phase2: 0.7418
• Lang Token Acc: 0.9995

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.98) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.05
• num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ach	Cer Ach	Zindi Ach	Wer Mas	Cer Mas	Zindi Mas	Wer Nyn	Cer Nyn	Zindi Nyn	Wer	Cer	Zindi	Zindi Strip	Zindi Phase2	Lang Token Acc
0.8792	0.1806	200	0.4402	0.3577	0.1403	0.7510	0.5068	0.1117	0.6907	0.3721	0.0863	0.7708	0.4147	0.1085	0.7384	0.7662	0.7375	0.9995
0.8004	0.3612	400	0.4317	0.3490	0.1380	0.7565	0.5052	0.1112	0.6918	0.3661	0.0851	0.7744	0.4094	0.1073	0.7416	0.7691	0.7409	0.9995
0.8415	0.5418	600	0.4408	0.3507	0.1373	0.7560	0.5079	0.1109	0.6906	0.3641	0.0853	0.7753	0.4102	0.1071	0.7413	0.7688	0.7406	0.9995
0.9093	0.7223	800	0.4306	0.3517	0.1424	0.7530	0.5053	0.1106	0.6921	0.3669	0.0853	0.7739	0.4105	0.1082	0.7407	0.7679	0.7396	0.9985
0.8229	0.9029	1000	0.4343	0.3548	0.1397	0.7527	0.5053	0.1106	0.6921	0.3657	0.0854	0.7745	0.4111	0.1076	0.7406	0.7679	0.7398	0.9995
0.7539	1.0831	1200	0.4420	0.3518	0.1396	0.7543	0.5069	0.1110	0.6910	0.3699	0.0866	0.7717	0.4121	0.1082	0.7398	0.7675	0.7390	0.9990
0.6852	1.2637	1400	0.4451	0.3497	0.1383	0.7560	0.5030	0.1099	0.6935	0.3670	0.0860	0.7735	0.4091	0.1073	0.7418	0.7690	0.7410	0.9995
0.6786	1.4442	1600	0.4390	0.3557	0.1401	0.7521	0.4994	0.1100	0.6953	0.3682	0.0865	0.7727	0.4102	0.1079	0.7410	0.7684	0.7400	0.9995
0.6951	1.6248	1800	0.4445	0.3464	0.1377	0.7579	0.5028	0.1100	0.6936	0.3651	0.0856	0.7747	0.4073	0.1070	0.7428	0.7701	0.7421	0.9995
0.6779	1.8054	2000	0.4408	0.3480	0.1378	0.7571	0.5007	0.1100	0.6946	0.3657	0.0858	0.7743	0.4074	0.1071	0.7428	0.7702	0.7420	0.9995
0.6474	1.9860	2200	0.4383	0.3483	0.1382	0.7568	0.5018	0.1096	0.6943	0.3658	0.0857	0.7742	0.4079	0.1070	0.7426	0.7700	0.7418	0.9995
0.6474	2.0	2216	0.4384	0.3489	0.1383	0.7564	0.5010	0.1094	0.6948	0.3660	0.0856	0.7742	0.4078	0.1069	0.7426	0.7700	0.7418	0.9995

Framework versions

• Transformers 5.14.1
• Pytorch 2.13.0+cu130
• Datasets 3.6.0
• Tokenizers 0.22.2