Bryant Lee issues

Repositories
Issues
Comments

Results 2 issues of


                                            Bryant Lee

Tokenizer saved with save_pretrained_merged(save_method = 'lora') does not match its tokenizer_config.json set

Hi~ I trained a LoRA model based on the `unsloth/mistral-7b-bnb-4bit` model. During the training, I modified the tokenizer's chat_template and eos_token using `get_chat_template()`. After the training was completed, I saved...

fixed - pending confirmation

Unable to achieve the score released by paper

I execute the source code and the parameters have not changed, but the score I got was only 11(BLEU-4). The score released by paper is 13.98(BLEU-4). I want to ask...