This is a brief introduction for model ckpts.

original models:

1. bigvgan_discriminator.pth
2. bigvgan_generator.pth
3. bpe.model (tokenizer)
4. config.yaml (original inference configuration)
5. gpt.pth (backbone transformer)
6. dvae.pth (use for getting training audio tokens)
7. unigram_12000.vocab (tokenizer vocabulary)

fine-tuned models:

1. Cantonese Model: a. config_yue.yaml b. gpt_tuned_yue.pth c. gpt_yue.pth (useless only for start stage in training)
2. Janpanese Model: a. config_ja.yaml b. gpt_tuned_ja.pth c. gpt_tuned_ja_v2.pth c. gpt_ja.pth (useless only for start stage in training)
3. Malaya Model: a. config_malay.yaml b. gpt_malaya_tuned.pth c. gpt_malay.pth (useless only for start stage in training)