what does --ckpt mean in eval script?

[hill2hill]

# model name and model path
--model_name
--model_path
# load model from ckpt
--ckpt

when should i set this param? its default value is "", but code only check whether it is None.

[hill2hill]

In my case, I just modified the --ckpt to be None. And it looks like working fine, i can reproduce the textVQA result with run_inference.sh. One thing might be helpful when somebody meet the error like: TypeError: MiniCPMV.chat() missing 1 required positional argument: 'image'. You can add one line arg like:

res = self.model.chat(
            image=image,
            msgs=msgs,
            context=None,
            tokenizer=self.tokenizer,
            **default_kwargs
        )

if using --generate_method interleave in run_inference.sh

[Cuiunbo]

thanks～