How to deal with the generate response is None, which the model directly generate the <im_end> token.

[unicorneeee]

I hope this message find you well. When I use the InfiniteBench to compute score, if the response of the model is None , it will raise an mistake,therefore, in our result table with the results of gpt4 and chatglm3 , etc.... Is these evaluate results remove the test example which response is None, and calculate the final score or just set all F1,Recall ,score to 0 for the cases which model's response is None, and calculate the final score?

[tuantuanzhang]

if a model generates none, it will get zero at this example