yuh-zha / AlignScore

ACL2023 - AlignScore, a metric for factual consistency evaluation.
https://arxiv.org/pdf/2305.16739.pdf
MIT License
113 stars 18 forks source link

Some weights of RobertaModel were not initialized #18

Open timbmg opened 6 months ago

timbmg commented 6 months ago

I have downloaded the AlignScore-base.ckpt and am initialzing AlignScore like so:

align_evaluator = AlignScore(
        model="roberta-base",
        batch_size=4,
        device="cpu",
        ckpt_path="/path/to/AlignScore-base.ckpt",
        evaluation_mode="nli",
)

However, this gives me the following warning, which concerns me a bit:

Some weights of RobertaModel were not initialized from the model checkpoint at roberta-base and are newly initialized: ['roberta.pooler.dense.bias', 'roberta.pooler.dense.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.

Is this expected? Will the evaluation be correct? I am assuming that these are not required because different heads are used, but woudl like to make sure.

maxyousif15 commented 6 months ago

I've had a similar warning, but I'm assuming it is safe to ignore from this issue, which seems identical, despite a different warning message?