You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

w2vbert-shona-waxal-punct

Dedicated Shona W2V-BERT CTC trained WITH punctuation in vocab, on WAXAL Shona.

Best validation:

Epoch	Step	Val Loss	WER keep	CER keep	Zindi keep	WER strip	Zindi strip
0.23	4	5.6245	0.9999	0.8569	0.0716	1.0063	0.0759
0.47	9	3.9539	1.0005	0.8681	0.0657	1.0005	0.0666
0.70	14	3.4340	1.0000	0.9657	0.0171	1.0000	0.0174
0.94	19	2.9095	1.0000	0.9941	0.0030	1.0000	0.0030
1.17	24	1.0690	0.9233	0.2838	0.3965	0.9200	0.4018
1.40	29	0.0668	0.1481	0.0228	0.9146	0.0941	0.9428
1.64	34	0.0486	0.1268	0.0185	0.9273	0.0767	0.9532
1.87	39	0.0424	0.1180	0.0169	0.9326	0.0685	0.9580
2.10	44	0.0402	0.1150	0.0162	0.9344	0.0639	0.9609
2.34	49	0.0366	0.1089	0.0150	0.9380	0.0610	0.9626
2.57	54	0.0346	0.1056	0.0142	0.9401	0.0597	0.9635
2.80	59	0.0329	0.1041	0.0140	0.9409	0.0582	0.9643
3.04	64	0.0344	0.1037	0.0138	0.9413	0.0577	0.9647
3.27	69	0.0345	0.1014	0.0139	0.9424	0.0575	0.9648
3.51	74	0.0345	0.1028	0.0143	0.9414	0.0572	0.9650
3.74	79	0.0348	0.1057	0.0151	0.9396	0.0577	0.9647
3.97	84	0.0326	0.1043	0.0141	0.9408	0.0584	0.9644
4.21	89	0.0363	0.1031	0.0143	0.9413	0.0576	0.9648
4.44	94	0.0356	0.1059	0.0152	0.9395	0.0586	0.9642
4.67	99	0.0332	0.1043	0.0155	0.9401	0.0582	0.9645
4.91	104	0.0332	0.1037	0.0156	0.9404	0.0573	0.9649
5.14	109	0.0352	0.1049	0.0143	0.9404	0.0579	0.9646

Safetensors

Model size

0.6B params

Tensor type

F32