w2vbert-shona-sd2

This model is a fine-tuned version of sulaimank/w2vbert-shona-waxal-punct-v2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0262
• Wer Keep: 0.1163
• Cer Keep: 0.0183
• Zindi Keep: 0.9327
• Wer Strip: 0.0318
• Zindi Strip: 0.9820
• Zindi Lower: 0.9977

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: 3e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• 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: 0.1
• num_epochs: 8.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Keep	Cer Keep	Zindi Keep	Wer Strip	Zindi Strip	Zindi Lower
3.4644	0.1754	200	0.3877	0.3874	0.0602	0.7762	0.3118	0.8204	0.8817
1.3694	0.3509	400	0.0587	0.1785	0.0243	0.8986	0.0828	0.9531	0.9911
0.8931	0.5263	600	0.0487	0.1595	0.0219	0.9093	0.0658	0.9627	0.9921
1.0625	0.7018	800	0.0458	0.1514	0.0205	0.9141	0.0598	0.9660	0.9924
1.5920	0.8772	1000	0.0457	0.1580	0.0208	0.9106	0.0605	0.9657	0.9930
0.9844	1.0526	1200	0.0408	0.1471	0.0196	0.9166	0.0533	0.9698	0.9947
0.4298	1.2281	1400	0.0397	0.1478	0.0201	0.9161	0.0535	0.9696	0.9928
1.1191	1.4035	1600	0.0403	0.1471	0.0204	0.9162	0.0531	0.9698	0.9925
0.7880	1.5789	1800	0.0380	0.1437	0.0193	0.9185	0.0519	0.9705	0.9939
0.9806	1.7544	2000	0.0376	0.1441	0.0192	0.9183	0.0518	0.9706	0.9940
0.7175	1.9298	2200	0.0369	0.1403	0.0185	0.9206	0.0476	0.9730	0.9948
0.8859	2.1053	2400	0.0376	0.1436	0.0206	0.9179	0.0502	0.9715	0.9943
0.1501	2.2807	2600	0.0363	0.1406	0.0189	0.9203	0.0476	0.9730	0.9948
0.9371	2.4561	2800	0.0362	0.1408	0.0189	0.9201	0.0498	0.9717	0.9943
0.5877	2.6316	3000	0.0354	0.1395	0.0184	0.9210	0.0482	0.9727	0.9944
0.7167	2.8070	3200	0.0346	0.1367	0.0182	0.9225	0.0452	0.9744	0.9957
0.3904	2.9825	3400	0.0348	0.1357	0.0187	0.9228	0.0463	0.9738	0.9946
0.5262	3.1579	3600	0.0345	0.1335	0.0177	0.9244	0.0434	0.9754	0.9961
0.3766	3.3333	3800	0.0342	0.1344	0.0182	0.9237	0.0436	0.9752	0.9958
0.2443	3.5088	4000	0.0335	0.1344	0.0215	0.9221	0.0441	0.9750	0.9961
0.5028	3.6842	4200	0.0324	0.1333	0.0183	0.9242	0.0431	0.9756	0.9963
0.4943	3.8596	4400	0.0336	0.1350	0.0184	0.9233	0.0432	0.9756	0.9964
0.8414	4.0351	4600	0.0334	0.1311	0.0204	0.9242	0.0421	0.9761	0.9960
0.3412	4.2105	4800	0.0318	0.1305	0.0221	0.9237	0.0414	0.9765	0.9964
0.2376	4.3860	5000	0.0338	0.1384	0.0195	0.9210	0.0469	0.9735	0.9964
0.2419	4.5614	5200	0.0314	0.1307	0.0194	0.9249	0.0401	0.9773	0.9962
0.7453	4.7368	5400	0.0306	0.1271	0.0184	0.9272	0.0390	0.9779	0.9964
0.4851	4.9123	5600	0.0304	0.1269	0.0195	0.9268	0.0389	0.9780	0.9969
0.4616	5.0877	5800	0.0312	0.1256	0.0190	0.9277	0.0385	0.9781	0.9964
0.2200	5.2632	6000	0.0298	0.1254	0.0185	0.9281	0.0381	0.9784	0.9967
0.3345	5.4386	6200	0.0293	0.1250	0.0198	0.9276	0.0369	0.9791	0.9970
0.5913	5.6140	6400	0.0290	0.1238	0.0195	0.9283	0.0361	0.9795	0.9970
0.2286	5.7895	6600	0.0295	0.1234	0.0182	0.9292	0.0358	0.9797	0.9969
0.0759	5.9649	6800	0.0283	0.1212	0.0176	0.9306	0.0344	0.9805	0.9972
0.0816	6.1404	7000	0.0283	0.1212	0.0187	0.9300	0.0343	0.9806	0.9973
0.4429	6.3158	7200	0.0279	0.1207	0.0187	0.9303	0.0342	0.9806	0.9974
0.1917	6.4912	7400	0.0273	0.1204	0.0196	0.9300	0.0333	0.9812	0.9974
0.4372	6.6667	7600	0.0269	0.1188	0.0190	0.9311	0.0329	0.9813	0.9975
0.3077	6.8421	7800	0.0273	0.1183	0.0185	0.9316	0.0327	0.9814	0.9976
0.3378	7.0175	8000	0.0270	0.1183	0.0189	0.9314	0.0324	0.9817	0.9976
0.0778	7.1930	8200	0.0269	0.1175	0.0183	0.9321	0.0323	0.9817	0.9977
0.6953	7.3684	8400	0.0265	0.1168	0.0188	0.9322	0.0313	0.9823	0.9979
0.4383	7.5439	8600	0.0265	0.1165	0.0184	0.9325	0.0318	0.9820	0.9977
0.4481	7.7193	8800	0.0262	0.1164	0.0187	0.9324	0.0316	0.9821	0.9977
0.3156	7.8947	9000	0.0262	0.1161	0.0183	0.9328	0.0317	0.9821	0.9977
0.3048	8.0	9120	0.0262	0.1163	0.0183	0.9327	0.0318	0.9820	0.9977

Framework versions

• Transformers 5.14.1
• Pytorch 2.13.0+cu130
• Datasets 5.0.1
• Tokenizers 0.22.2