The API / playground feedback results are inconsistent

API
1

image
image2840×974 146 KB

thanks for help.
1.Using the same prompt word, the API feedback did not recognize the negative sign, but the playground did. How to correct it?
2.prompt “Identify the problem, note - it may be a negative sign”
3.API feedback:

\[\begin{cases}
\frac{1}{x+y} - \frac{3}{x-y} = \frac{1}{2} \\
\frac{2}{x+y} + \frac{1}{x-y} = \frac{4}{3}
\end{cases}\]

\begin{cases} \frac{1}{x+y} - \frac{3}{x-y} = \frac{1}{2} \\ \frac{2}{x+y} + \frac{1}{x-y} = \frac{4}{3} \end{cases}

it’s should be “-1/2”

2

Are you sending the exact save call to the API as you are in the playground?

You can view the code for the playground request and just send that to test.

Also, remember the models are probabilistic, your need to send the same request in the playground and in the API probably close to 100 times each to discern any actual difference in the quality of responses.

3

1.yes , exact same
2.“You can view the code for the playground request and just send that to test.” I am a beginner, can you please give me some guidance on how to use it?

4

1000019848
10000198481080×2221 206 KB

1000019849
10000198491080×2222 180 KB

This way you can ensure the requests are exactly the same.

But, like I wrote earlier, there’s a lot of randomness, so you’ll need to make many requests to determine if there’s any actual difference in the quality of results or if you just had bad luck on one of the generations.

5

thanks for help. I will try it

1 Like