It seems that the data format of aplaca.json does not conform to the read logic of dataLoader.

[hljjjmssyh]

[Linear95]

My Bad. You need to use the prompts from the Standford alpaca repo to get 'query' and 'target'.

[hljjjmssyh]

When obtaining the "target", do we need to add "tokenizer.eos"?
https://github.com/tatsu-lab/stanford_alpaca/blob/761dc5bfbdeeffa89b8bff5d038781a4055f796a/train.py#L141

[Linear95]

You dont need to. We will automatically add eos token in our data collactor.