fine-tuning on CausulLM or SequenceClassification

#3
by iman1995 - opened

How do you fine-tuned the model. the config file contains CausalLM version, but the target task is a sequence classification.

Bespoke Labs org

Hi Iman, the model is fine-tuned to generate output text as either 'Yes' or 'No'. We are using the probability of the model in generating the first token as 'Yes'. At inference time, the default threshold value is set to 0.5. If the probability that the first generated token is 'Yes' is greater than 0.5, we classify the claim as supported by the document.

Sign up or log in to comment