jed351/gpt2-base-zh-hk

This model has not been trained on any Cantonese material.

It is simply a base model in which the embeddings and tokenizer were patched with Cantonese characters. One can find the original model gpt2-tiny-chinese.

I used this repo to identify missing Cantonese characters

My forked and modified version

After identifying the missing characters, the Huggingface library provides very high level API to modify the tokenizer and embeddings.

Download a tokenizer and a model from the Huggingface library. Then: 

tokenizer.add_tokens("your new tokens")
model.resize_token_embeddings(len(tokenizer))

tokenizer.push_to_hub("your model name")