I am trying to do some special prompting methods on a dataset called Winoground (just providing for context, not really relevant to the error itself). But I am getting a 400 error for a valid Python String. It can even be printed with no issues.
Blockquote""\nSpecifically, you are a neutral expert evaluating the responses of two other experts, one who believes the correct answer is A and the other who believes the correct answer is B.\n\nGiven the four explanation, pick the one that is correct (1, 2, 3, or 4) and return the right answer (you must choose either A or B even if unclear).\n\nReasoning: 1In the scene graph, the relationship predicate indicates that the subject, which is object 1 (grandfather), is kissing object 2 (girl), correlating with the scenario in caption A where ‘an old person kisses a young person’. Additionally, the spatial location of the objects also supports this explanation as the grandfather is located lower than the girl, which is usually the spatial arrangement when an individual is kissing another on the forehead or cheek.2Examining the attributes provided in the scene graph, object 1 (grandfather) is described as ‘old’ and object 2 (girl) is labeled as ‘daughter’, suggesting a younger age in contrast to the grandfather. This age distinction aligns with caption A ‘an old person kisses a young person’ and signifies a usual family situation where an elder kisses a younger member. This supports the argument that caption A is the correct interpretation of the scene graph.\n\nResponse Format Example:\nCorrect Reasoning:\n3\n\nAnswer:\nB\n" is not of type ‘array’ - ‘messages’"
is the message. Yes, other prompts are working fine, so it isn’t an API issue. I know the following tool is only an estimate, but it definitely isn’t a token limit issue either:
Would appreciate any insight. Thanks!
Thank you for pointing out that part of the error. I did not notice that my prompt setup for that specific query was incorrectly formatted. Fixed!