Can you please share more about your process
#1
by
macadeliccc
- opened
What dataset did you use to fine tune?
Did you filter out the samples that were used in gsm8k train data? I believe this model is contaminated with gsm8k train data making it’s scores invalid
Yes, the dataset was annotated with if rows were in Gsm8k and I filtered them out before training using the provided scoring filter in the repo
macadeliccc
changed discussion status to
closed
Hey @SanjiWatsuki ,
Can you say what the base model was?