stefan-it/hmbench-topres19th-en-hmbert_64k-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-2

Fine-tuned Flair Model on TopRes19th English NER Dataset (HIPE-2022)

This Flair model was fine-tuned on the TopRes19th English NER Dataset using hmBERT 64k as backbone LM.

The TopRes19th dataset consists of NE-annotated historical English newspaper articles from 19C.

The following NEs were annotated: BUILDING, LOC and STREET.

Results

We performed a hyper-parameter search over the following parameters with 5 different seeds per configuration:

• Batch Sizes: [4, 8]
• Learning Rates: [3e-05, 5e-05]

And report micro F1-score on development set:

Configuration	Seed 1	Seed 2	Seed 3	Seed 4	Seed 5	Average
bs8-e10-lr5e-05	0.7918	0.7984	0.786	0.7841	0.7992	0.7919 ± 0.0069
bs8-e10-lr3e-05	0.7886	0.8142	0.7925	0.7865	0.7757	0.7915 ± 0.0141
bs4-e10-lr3e-05	0.7838	0.7885	0.7934	0.8049	0.7862	0.7914 ± 0.0084
bs4-e10-lr5e-05	0.7621	0.7017	0.7578	0.7708	0.7686	0.7522 ± 0.0287

The training log and TensorBoard logs (not available for hmBERT Base model) are also uploaded to the model hub.

More information about fine-tuning can be found here.

Acknowledgements

We thank Luisa März, Katharina Schmid and Erion Çano for their fruitful discussions about Historic Language Models.

Research supported with Cloud TPUs from Google's TPU Research Cloud (TRC). Many Thanks for providing access to the TPUs ❤️